{"report":{"v":1,"package":"@sellable/mcp","version":"0.1.1020-wip.morningrefill.20261113.1","ranAt":"2026-11-16T11:32:29.064Z","install":{"ok":true,"bin":"mcp","integrity":"sha512-MKbkKRzyM/TDEkoAKirahl4hdTH9Vn6iR9HoL/KPdtZkbtkToeN4ML5TmRHaGQG1NabRTE/Sd1rCLHNWft8HQA==","deps":96,"scripts":[],"bins":["mcp","sellable-mcp","sellable-mcp-dev"],"license":"MIT"},"start":{"ok":true,"ms":265,"serverName":"sellable-mcp","serverVersion":"0.1.1020-wip.morningrefill.20261113.1","protocol":"2025-06-18","stderr":"[sellable-mcp] pid=96 starting...\n[sellable-mcp] pid=96 ready in 26ms (tools=203)\n"},"tools":[{"name":"list_agent_endpoints","description":"List the current Agent's stable HTTP endpoints and their secret-free publication metadata.","params":[]},{"name":"publish_agent_endpoint","description":"Atomically create a stable endpoint from slug and name, or update an existing endpointId, and publish one immutable prompt/schema revision. Always pass the required fields mainPrompt, inputSchema, and outputSchema; the prompt field is named mainPrompt, not prompt. timeoutSeconds defaults to 600 when omitted. If the request names an immutable timeout, pass that exact timeoutSeconds in the same atomic publish call and verify the returned or listed immutable revision before claiming success. Public invocation input is native text (or a supported text-message array), never an object such as {\"text\":\"...\"}; for a text-only endpoint use inputSchema {\"type\":\"string\",\"minLength\":1} and curl body {\"mode\":\"async\",\"input\":\"Your request\"}. Supported JSON Schema is intentionally strict: every schema node must use one string-valued \"type\" from object, array, string, number, integer, boolean, or null. Object schemas require properties plus additionalProperties:false; array schemas require items. Composition keywords and array-valued nullable types are unsupported; omit an optional property instead of using nullable unions. The Agent and workspace are runtime-bound, never caller-selected.","params":["endpointId","slug","name","mainPrompt","inputSchema","outputSchema","timeoutSeconds","concurrency"]},{"name":"rotate_agent_endpoint_key","description":"Reveal the initial endpoint invocation key or rotate the current credential with no overlap. Plaintext is returned exactly once in the reveal field.","params":["endpointId","oldCredentialVersionId","nextGeneration","expiresAt"]},{"name":"list_agent_endpoint_runs","description":"List conversations for one endpoint, or the bounded run summary for one conversation.","params":["endpointId","conversationId"]},{"name":"get_agent_endpoint_run","description":"Read one sanitized run detail with bounded events, terminal result, attempts, proposals, decisions, effect intents, and receipts.","params":["endpointId","conversationId","runId","afterSequence"]},{"name":"decide_agent_endpoint_proposal","description":"Approve exactly once or deny one current endpoint proposal using its exact fingerprint and fence. Persistent/session approval is not supported.","params":["endpointId","conversationId","runId","attemptId","proposalId","proposalFingerprint","fence","decision"]},{"name":"stop_agent_endpoint_run","description":"Request stop for one exact endpoint run and return the durable terminal or uncertain receipt.","params":["endpointId","conversationId","runId","expectedProposalFingerprint"]},{"name":"validate_campaign1_kickoff_handoff","description":"Read-only validation for exact inline sellable.kickoff.campaign1.v1 Markdown plus its Admin result binding. Returns sufficient Campaign 1 intake and campaign-brief Markdown without reading a URL, creating a campaign, sourcing/importing leads, or advancing approval gates.","params":["handoffMarkdown","binding","expected"]},{"name":"archive_campaign","description":"Archive a campaign through Sellable's product archive endpoint. This is an explicit cleanup/archive action, is idempotent, and may unschedule pending cells owned by that campaign table. Do not use from refill/setup commands unless the user explicitly asked to archive exact campaign ids.","params":["campaignId","workspaceId"]},{"name":"duplicate_campaign","description":"Duplicate a campaign by calling Sellable's existing duplicate endpoint. Optionally pass name to set the new CampaignOffer and workflow table name. Returns campaignOfferId, campaignName, and workflowTableId for the new copy.","params":["campaignId","name"]},{"name":"get_campaigns","description":"List campaigns for the authenticated user. Returns id, name, createdAt. Ordered by most recent first.","params":["limit"]},{"name":"get_campaign","description":"Get detailed campaign context including brief, positioning, sender info, lead scoring rubrics, stats, and follow-up automation settings. Returns workflowTableId for fetching rows. writers says who writes this campaign's first messages, replies and follow-ups (null = Sellable); writerEndpoints lists the endpoints that can be chosen.","params":["campaignId"]},{"name":"get_campaign_messages_preview","description":"Resolve campaign workflow table and return table meta + lead preview for messaging selection (supports filters + pagination).","params":["campaignId","tableId","leadLimit","page","filters"]},{"name":"create_campaign","description":"Create a new campaign offer OR resume an existing one. Low-level write tool: load create-campaign workflow instructions first via get_subskill_prompt({ subskillName: \"create-campaign\" }) (or bootstrap_create_campaign + nextStep). If campaignId is provided, this tool returns the watchUrl + state for that campaign instead of creating a new campaign.\n\nINPUTS:\n- Pass either `clientProspectId` (preferred) or `senderLinkedinUrl` for net-new campaigns.\n- `clientProspectId` is the EnrichedProspect row ID for the campaign identity/client prospect, used by ICP scoring, message generation, and brief rendering.\n- `senderLinkedinUrl` accepts a full LinkedIn profile URL or a public profile handle (for example, csreyes92) when `clientProspectId` is not yet materialized; the backend will attempt to resolve a campaign identity prospect from it.\n\nPREREQUISITES:\n- If you provide `senderLinkedinUrl`, it is used as a campaign identity bootstrap signal, not as a connected-sender availability check.\n- `clientProspectId` is optional when sender URL is available and allows a direct materialized campaign identity context.\n- `clientProspectId` is preferred for deterministic campaign identity resolution.","params":["campaignId","name","clientProspectId","senderLinkedinUrl","offerPositioning","campaignBrief","messageGenerationMode","messageApproach","currentStep","watchNarration","leadSourceType","leadSourceProvider","selectedLeadListId","senderIds"]},{"name":"update_campaign","description":"Update an existing campaign offer with positioning, brief, workflow state, or follow-up automation fields. Returns currentStep plus navigation diagnostics and persisted automation readback when automation settings change. Writers (openerEndpointId, replyEndpointId, followUpEndpointId) are chosen per campaign; Sellable writes anything left null.","params":["campaignId","workspaceId","name","offerPositioning","campaignBrief","dos","donts","headlineICPCriteria","leadSourceType","leadSourceProvider","selectedLeadListId","currentStep","currentStepTransition","clearCurrentStepIfMatches","watchNarration","interactionMode","enableICPFilters","useMessagingTemplate","messageApproach","autoApproveFollowups","enableFollowUps","autoApproveReplies","openerEndpointId","replyEndpointId","followUpEndpointId","senderIds","rubric","flowVersion"]},{"name":"start_campaign","description":"Start a paused campaign, enabling the sweeper to send messages. Call it when the customer explicitly asks to start, or when their standing consent covers it (customer_program context launchConsent.mode is autopilot, or no answer is saved; only review_first keeps the Start theirs). A sender the customer stopped (customer_program context stoppedSenderIds) comes off the campaign before it starts; the result names them in stoppedSendersRemoved. It must not be used as part of fill-send-horizon, message preparation, or scheduling-proof flows.","params":["campaignId","workspaceId"]},{"name":"pause_campaign","description":"Pause an active campaign, stopping the sweeper from sending messages. Pause only when the customer asks to stop a campaign. To stop old copy, switch the message (queue_campaign_cells rewrite); a pause is the customer's and stays until they ask to start it.","params":["campaignId","workspaceId"]},{"name":"update_campaign_brief","description":"Update the campaign brief markdown document. This is a living canvas where you accumulate knowledge during crafting. Use it to: add social proof/case studies discovered during research, note common objections, record successful hooks, or reorganize when user changes direction. You control the full markdown content - read the current brief first, then update it intelligently.","params":["campaignId","campaignBrief"]},{"name":"prepare_campaign_ab_test","description":"Prepare a campaign A/B test from a full Campaign A created through the Sellable create-campaign workflow. The helper duplicates the source campaign through the product duplicate path so B preserves the approved brief, ICP filters, message setup, and campaign-table structure before applying the B delta. sourceLeadListId is optional and must only be supplied after asking whether to split Campaign A's clean selected lead list. The tool never launches either campaign.","params":["sourceCampaignId","sourceLeadListId","variantName","variantBriefDelta","variantALabel","variantABriefDelta","splitStrategy","targetCounts","dryRun","idempotencyKey"]},{"name":"resolve_campaign_fill_route","description":"Audit-only resolver to call before interpreting plain \"fill campaigns\", \"load everyone up\", or similar fill requests. It decides whether the workspace has evergreen/horizon targets, current dashboard-active campaign-backed sequence targets, start-eligible PAUSED targets, or no targets so the agent must ask what to create. It does not create, append, prepare, approve, schedule, launch, archive, delete, send, or call lower-level fill tools. Only call fill_campaign_horizon when this resolver returns route \"evergreen_horizon\" and the selected returned target is eligible; use start_campaign only for exact returned startEligible PAUSED targets after the refill workflow approval/reread packet proves starting is needed. Exact target narrowing uses campaignId or tableId only; never fuzzy campaign names.","params":["intent","campaignId","tableId","limit","workspaceId"]},{"name":"get_campaign_refill_state","description":"read-only refill research primitive to call after resolve_campaign_fill_route and before any source import, message preparation, approval, scheduling, or horizon fill decision. It returns current campaign/table/source/sender/funnel/scheduler diagnostics, preparationFrontier for earliest unenriched rows and later prepared islands, plus freshness state for one exact campaignId or tableId only. This tool does not create rows, does not import leads, does not prepare messages, does not approve messages, does not schedule sends, does not launch campaigns, and does not expose direct campaign types as refillable targets. Exact targeting uses campaignId or tableId only; never pass campaign names or table names. sourceHistory lists what this campaign already tried: its provider searches (filters, import jobs, leads imported) and its Signal Discovery keywords with their selected posts. Pass counts are per campaign (counts.passed), not per search.","params":["campaignId","tableId","workspaceId","reportingContext"]},{"name":"get_refill_target_plan","description":"V1-ROUTE PLANNER — CHECK THE ROUTE BEFORE CALLING. If this server also exposes `refill_v3_advance`, that tool owns refill execution and makes its own fresh observation, so a refill must NOT be opened here; send every ordinary refill, top up, fill, max out, or load up request to `refill_v3_advance` instead. Call this tool only on servers where `refill_v3_advance` is absent. read-only refill target planner to call before any refill mutation. It identifies eligible senders and infers one implicit refill lane per sender. For unified Sales Nav cascade campaigns, the public refill lane is the paid-InMail cascade target (send_inmail_closed) with campaign classification sales_nav_cascade; the same campaign can still route prospects to Open InMail or connection-request fallback based on row eligibility and fresh paid credits >= 5. Non-cascade lanes remain connection invites or paid InMails, chosen from future scheduled, recent scheduled, and ready-to-schedule evidence in active campaign-backed sequence campaigns before falling back to campaign sequences. DMs remain follow-up sequence actions, not refill target lanes. By default the planner computes the scheduler-forward 48-hour target window, selected sender-local days whose sending windows overlap that window, gross target, actual sent coverage, scheduler-owned scheduled coverage, projected coverage (sent + scheduled), ready buffer, remaining projected gap, paid InMail credit/threshold/freshness feasibility, same-campaign connection fallback availability, bounded action candidates, and targetRevision drift proof. This tool does not create rows, import leads, prepare messages, approve messages, does not schedule sends, start campaigns, lower paid InMail thresholds, create campaigns, launch, spend InMail credits, or write scheduler fields. Complete projected targets no-op without approval; loaded, awaiting scheduler targets use read-only wait/reread until projected coverage fills.","params":["intent","horizonSendDays","untilDate","targetDate","campaignId","tableId","senderIds","senderNames","senders","actionTypes","approvalMode","workspaceId","targetConfig","runHandle","reportingContext","messageTemplateRevision"]},{"name":"get_scheduler_fill_capacity","description":"read-only scheduler capacity query for refill-sends before any refill mutation. It asks the product scheduler model how many additional cells the scheduler will try to place for exact sender/action requests, using the default scheduler-forward horizon or one exact targetDate. It returns scheduler-fillable slots, occupied scheduled/processing slots, rolling-weekly candidate timing that shows when capacity frees later in the send window, sender sendability gates, cooldowns, Sales Navigator status for paid InMail, cached paid-InMail credit feasibility, threshold source, blockers, warnings, and explicit sideEffects false. This tool does not create rows, import leads, prepare messages, approve messages, does not schedule sends, does not refresh paid InMail credits, mutate thresholds, start campaigns, launch, or send. Use exact senderId values only; never pass sender names.","params":["capacityRequests","targetDate","mode","workspaceId"]},{"name":"run_scheduler_sweep","description":"Trigger the product scheduler placement sweep for one explicit workspace now, or read the last on-demand scheduler run status with action \"status\". A run is workspace-wide, not a scoped run for one campaign/table. Small sweeps may return a terminal ran/window_closed_noop/failed receipt immediately; larger sweeps return attached with no receipt and retryAfterMs while the exact request scope continues durably, so poll action:\"status\" with the identical scope until its receipt appears and never redispatch it. Backoff replays can also return retryAfterMs. In v3 receipts, readyCellsFound is total scheduler-ready supply found before filters, cellsConsidered is the allocation-attempt count that survived prefilters, prefiltered means ready cells removed before allocation such as no capacity, stale paid InMail credit facts, or sender mismatch, skipped means considered cells rejected by hard scheduler gates, and deferred means considered cells waiting on windows/capacity/cooldown. campaignScopeSummary remains bounded display evidence, while expectedTargetSummary explains every hash-bound planner target exactly once without filtering placement. Expected targets require a frozen numeric total placement cap; unrelated eligible campaigns may consume that cap through normal workspace-wide gates, and changedCounts audits every actual placement by campaign, table, action, and sender. The receipt can include readyCellsByType, consideredCellsByType, prefilterReasons, skippedReasons, deferredReasons, summary, and nextAction. It places cells within existing scheduler gates only: it can move placement earlier, but it cannot bypass sending windows, daily limits, cooldowns, sender gates, billing, or credit thresholds, and it never sends messages directly. Repeated calls inside the backoff window replay the last receipt verbatim. Status is read-only and scoped to the last on-demand run only; cron sweeps are not recorded here.","params":["workspaceId","action","targetDate","requestKey","targetShapeRevision","expectedTargets","expectedTargetsHash","maxPlacements"]},{"name":"refill_sends","description":"SUPERSEDED V1 REFILL COORDINATOR — CHECK THE ROUTE BEFORE CALLING. If this server also exposes `refill_v3_advance`, that tool owns refill execution and this one must NOT be called: every ordinary request to refill, top up, fill, max out, or load up sends goes to `refill_v3_advance` with exactly {workspaceId, scope}. Call this tool only on servers where `refill_v3_advance` is absent. Typed command entrypoint for Sellable refill sends. Refill defaults to yolo:true; pass explicit yolo:false only for a review-first/manual run. Every mode begins with a read-only visible plan pinned to expectedTargetShapeRevision and expectedActionKey; no credit refresh, refill mutation, or run fence may exist before that render. A first manual call without those pins returns status plan_ready and requires explicit acceptance before its planned_manual_run continuation. A first default workspace-wide yolo coordinator call without pins also returns status plan_ready with the fresh sender/campaign waterfall, selected exact target, and a planned_workspace_run continuation; yolo auto-accepts the displayed packet, then traverses its planner-ranked existing targets autonomously. Execution preflights globalActionQueue[0] against the rendered authority and executes one fenced checkpointed run slice before rereading workspace scope for the next exact target. An active_exact_run continuation carries targetConfig plus exact runId/fence; a next_exact_target continuation carries the next preflighted targetConfig and opens a new fence. Exact target configs bind the issued semantic action and revisions. After a terminal exact run, two fresh read-only preflights stabilize the workspace queue; if the same terminal action remains visible, the coordinator excludes only that exact sender/lane edge and preflights the next planner-ranked existing edge instead of reopening an identical fence. A retryable Signal Discovery provider outage defers that exact signal-search action family within its campaign/table/lane for the remainder of the current workspace command, even if the eligible sender cohort changes, records the blocker in workspaceRunState, and continues independent scheduler, credit, or source-family work; a fresh user command retries the deferred family. If no other exact edge exists while the workspace gap remains positive, it returns the concrete deferred provider blocker or a classified typed terminal: new_campaign_required when receipts prove every existing lane/source family is spent, actionable_supply_not_queued (retryable) when actionable row frontiers remain but no executable exact edge surfaced, or workspace_exact_target_unavailable for the residual case. Once a scheduler sweep is dispatched, the exact fence remains active and every in_progress continuation polls only that request key until its official terminal receipt. A host wait budget never converts the active sweep into loaded_awaiting_scheduler. The post-receipt target reread returns complete, another planner action, ","params":["yolo","executionMode","requireWorkspace","workspaceId","senders","senderIds","senderNames","actionTypes","horizonSendDays","untilDate","targetDate","campaignId","tableId","intent","approvalMode","expectedTargetShapeRevision","expectedActionKey","runId","fence","runHandle","targetConfig","reportingContext","messageTemplateRevision","agentKeywords","workspaceCoordinator","workspaceRunState","backgroundWait"]},{"name":"refill_v3_world_state","description":"One factual observation of Refill V3 send capacity. It returns one row per scoped sender, per scoped local date, per lane, where the two lanes connection_invite and paid_inmail are independent: each carries its own slots, its own typed blocker, and its own eligible and ineligible campaign options. Each campaign's facts (name, status, supply, evidence) appear once in top-level `campaigns`; a row's `activeCampaigns` and `inactiveCampaigns` name those campaigns by id with that row's own eligibility and reason. A scoped sender/date request refreshes both lanes of that sender and date atomically, so the two lanes are never read from different instants. It states facts and never chooses: it schedules nothing, places nothing, sends nothing, approves nothing, enriches nothing, generates nothing, and mutates no campaign, and it reports each of those explicitly as false. The one bookkeeping effect it may perform is a paid-InMail credit cache read or refresh, reported as its own outcome.","params":["workspaceId","campaignId","scope"]},{"name":"refill_v3_fill_ready","description":"Place just-in-time ready outreach for exactly one workspace, sender, sender-local date, lane, and campaign. It MUTATES product state: within the exact bounded cohort it approves prepared rows that still need approval, assigns sender affinity, and schedules ready actions through the on-demand scheduler. The two lanes connection_invite and paid_inmail are independent and a paid request re-attests paid-InMail credits immediately before placement. It never creates rows, imports leads, generates messages, starts or launches a campaign, changes thresholds, or sends anything directly. Every attempt is bounded by the required idempotencyKey: reuse the SAME idempotencyKey to retry one request, and expect a closed terminal outcome of scheduled, partially_scheduled, awaiting_scheduler, campaign_exhausted, or blocked, each with a typed reason and named clamp. An awaiting_scheduler answer carries a pollKey whose only valid follow-up is a status lookup, never a second placement.","params":["workspaceId","senderId","targetDate","lane","campaignId","requestedCount","idempotencyKey"]},{"name":"refill_v3_edit_selected_campaign","description":"Apply exactly ONE allowlisted configuration, filter, or source field patch to exactly one already-selected campaign, proven by the exact bounded evidence that chose it. It MUTATES campaign configuration and is deliberately narrow: it wraps and restricts update_campaign and refuses the campaign name, brief, positioning, senders, current step, interaction mode, provider or list replacement, rubric, template or sequence meaning, and any launch or start. It never reruns cells, regenerates messages, relaxes a rubric, writes raw rows, schedules, or sends; those keep their existing owners. A filter or source value that weakens the approved objective or any employment, DNC, legal, or explicit-exclusion invariant is invalid at the schema and makes zero delegated calls. Connection-health facts are categorically not edit actions. Every success returns immutable before/after fingerprints, the exact changed fields, a deterministic effect key, replay state, rollback data, and a required exact sender/date two-lane re-observation; reuse the same idempotencyKey to retry one request.","params":["workspaceId","campaignId","tableId","campaignVersion","evidenceFingerprint","priorActionFingerprint","action","field","value","reobservationScope","idempotencyKey"]},{"name":"refill_v3_advance","description":"THE DEFAULT REFILL ENTRY POINT. Whenever this tool is exposed, every ordinary request to refill, top up, fill, max out, load up, or otherwise add sends for a workspace, a sender, or a day routes HERE, and `refill_sends` is the superseded V1 fallback that applies only to servers which do NOT expose this tool. Never infer the route: use tool availability. When the operator names a workspace, resolve it with `list_workspaces`; fail closed if no accessible workspace matches or more than one matches, and never mutate the active workspace. Use configured context only when the operator named none. Start here: do not open the refill ladder with `get_refill_target_plan` or `refill_sends`. Advance exactly ONE fresh, fenced Refill V3 step for one workspace and scope. Pass only {workspaceId, scope}; refill defaults to yolo:true. YOLO is an autonomy grant, not a mode. Pass explicit yolo:false only when the operator asks for a conservative, review-first run. MCP carries signed run continuity internally. An advanced result is ALREADY EXECUTED and has reread its authority. For campaign_attention_required, call refill_v3_continue first with no decision: live-job waiting, exact actionable enrichment, and same-template generation are deterministic. Only model_decision_required routes the returned accurate campaign packet to refill-sends-work-campaign for one repair, criteria, rubric, waterfall, or source-expansion judgment. Repeat until complete, awaiting_external_change, or blocked, and report every terminal sender/date/lane row. A thrown error is not a terminal: call this tool once more with the identical workspaceId and scope (the MCP keeps the run current) and report a blocker only if it fails the same way.","params":["workspaceId","campaignId","scope","yolo"]},{"name":"refill_v3_continue","description":"Continue exactly one server-issued Refill V3 campaign-attention packet. continuation_pending means the same operation is still running: wait retryAfterMs and call this tool with identical scope and no decision until it returns. Reconcile transport timeouts the same way; do not call advance or dispatch another search while pending. Call first without decision: after a fresh revalidation, the tool deterministically owns live-job waiting, selected-source row copying, exact actionable enrichment, and same-template message generation. It returns model_decision_required only when the accurate packet needs a genuine repair, criteria, rubric, waterfall, or source-expansion judgment; then call again with that one closed decision. Only an exact fresh packet can dispatch one existing campaign-work authority, which re-observes the exact sender/date and settles before a sibling campaign may advance. The caller cannot supply batch sizes, readiness digests, retry counters, fill inputs, scheduler inputs, or idempotency keys. A result with recoveredFromError already re-observed after a failure: handle it by its kind. A thrown error is not a terminal: call refill_v3_advance once more with the identical scope and report a blocker only if it fails the same way.","params":["workspaceId","scope","campaignId","yolo","decision"]},{"name":"setup_evergreen_campaigns","description":"Evergreen campaign setup plan/verify command. Use plan mode first to inspect exact workspace/sender/campaign/table/source state and receive immutable lane packets. `selectedSenderIds` is a legacy shorthand for both scopes; prefer `postEngagerSenderIds` for the Post Engagers sender scope and `sharedSenderIds` for the shared lane sender scope when they differ. `sharedSenderIds` must include every intended shared-lane sender, not only the named Post Engagers sender; when adding other senders to existing shared evergreen campaigns, resolve those sender ids with list_senders and pass the full shared set. If a protected existing Post Engagers campaign must stay unchanged and the operator requested only shared lane execution, pass postEngagerSenderIds:[] and sharedSenderIds; do not include the protected active Post Engagers lane as a reuse packet because it can make the yolo plan non-autoExecutable. The command plans one Post Engagers lane per post-engager sender plus shared Signal Discovery and Shared Cold Fallback lanes for the shared sender set. Non-archived PAUSED/ACTIVE/DRAFT shared campaigns/tables are the reusable targets; archived campaigns, archived tables, and archived waterfall bindings are stale inventory and must not be counted as valid evergreen slots. If an existing non-archived shared campaign is already attached to additional senders, plan reporting should count those attached senders as existing shared-lane membership rather than treating them as missing or duplicate slots. yolo is only a parent-skill auto-execution hint for safe lane packets; pass yolo only in plan mode and never include yolo on mode:\"verify\" calls. This backend command remains read-only in plan mode and verifies receipts in verify mode. Package-backed prompt authority: the installed public wrapper can be the local Codex skill entrypoint, but lane workers must use get_subskill_prompt and get_subskill_asset for nested `$sellable:create-campaign`, create-campaign-v2, generate-messages, validation, and assets; nested filesystem prompt fallback is a failed UAT. Use mcp__sellable only for workspace selection and product mutations/readbacks. Do not use mcp__sellable_admin, direct DB, Prisma, SQL, built-in web search, browser search, web.run, or any external browsing/search tool as execution or research proof; if Sellable MCP research tools are insufficient, write a blocked receipt instead of browsing externally. Worker-local replans are read-only drift checks and must preserve the exact parent sender scopes, including postEngagerSenderIds:[] when intentionally empty and the exact sharedSenderIds array; if scope, planRevision, actionId, or laneKey drifts, stop with blocked:worker_plan_scope_drift before mutation. Each lane packet includes workerDispatch with preferredRuntime, runtimeFallbackOrder, acceptedRuntimes, rejectedRuntimes, requiresVisibleThreadOrDurableReceipt, receiptArtifactHint, receiptRunId, and receiptMustBeWrittenAfter; pre-existing receipts at old determinis","params":["mode","workspaceId","campaignSequenceOptions","depth","handoffMode","yolo","allConnectedSenders","selectedSenderIds","postEngagerSenderIds","sharedSenderIds","bindings","planRevision","selectedActionIds","receipts"]},{"name":"fill_campaign_horizon","description":"Evergreen/horizon-only tool: audit or apply a bounded CampaignOffer horizon fill from Signal Discovery/source lead-list rows only when resolve_campaign_fill_route returns route evergreen_horizon or the user explicitly requested evergreen/horizon and the target is managed-waterfall eligible. Use audit first to get stateRevision, then apply with that stateRevision. Apply imports at most 300 eligible non-excluded source rows, starts bounded message preparation in approval mode, skips rows from excluded posts/authors, and does not start or launch the campaign, send messages, or directly assign scheduler-owned send timestamps. Prepared/approved rows are intermediate only; report scheduled completion only after re-reading scheduler-owned scheduled cells with non-null scheduler timestamps.","params":["action","campaignId","tableId","stateRevision","excludedPostIds","excludedPostUrls","excludedAuthorProfileUrls","excludedAuthorNames","targetPreparedMessages","maxRowsToCheck","batchSize","workspaceId"]},{"name":"start_campaign_message_preparation","description":"Start a bounded message-preparation job for a specific existing CampaignOffer campaignId/tableId. This is a shared execution primitive used by campaign setup and the fenced Refill V3 coordinator; for fill/refill-sends intent, call refill_v3_advance instead of manually reconstructing the loop. Compatibility/recovery callers may use it only after resolve_campaign_fill_route, an exact target re-read, and an active prep-job check. It is not campaign creation or evergreen horizon fill, and it never launches the campaign, sends messages, directly writes scheduledFor, or rewrites scheduled/attempted outreach. The job processes eligible rows top-down by WorkflowTableRow.position, queues pending Enrich Prospect cells first, lets ICP/rubric and Generate Message cascade, then marks ready or approves only the bounded cohort. needsGeneratedMessage is only for already-passed generated-message repair and can skip earlier unenriched rows. Prepared/approved/ready rows are intermediate only; scheduled completion requires a later re-read proving scheduler-owned scheduled cells with non-null scheduledFor. Surface active preparation jobs, exhausted source rows, disconnected Sales Nav/deleted sender accounts, missing sequence state, and other sender-health blockers separately from prepared/approved/scheduled counts. Omit maxRowsToCheck and batchSize for the adaptive default: calibrate on at least 100 actually-enriched rows, estimate the row budget from observed rubric/pass yield, cap rows at 300, and process at most 100 newly checked rows at a time. The worker will not pull another row batch while the current checked batch still has queueable or active cells. Do not interpret checkedRows as enriched rows; use progress.enrichedRows, needsEnrichRows, activeCellCount, preparedMessages, approvedMessages, and stopReason. Cross-workspace Agent calls must pass workspaceId.","params":["campaignId","tableId","targetPreparedMessages","maxRowsToCheck","batchSize","maxBatchRows","approvalMode","rowSelector","excludeRowIds","autoContinue","senderId","actionType","profileSchemaVersion","compilerVersion","sequenceVersion","campaignDigest","actionColumnId","targetColumnType","pathDigest","targetAction","targetNodeId","entryNodeId","pathNodeIds","dependsOnColumnIds","runCondition","intrinsicRequirements","approvalPolicy","requestId","effectId","disableLowPassRateStop","workspaceId","reportingContext","history"]},{"name":"get_campaign_message_preparation_status","description":"Poll compact campaign message preparation job progress: checked rows, pass/prepared/approved count, target, estimated row budget, adaptive batch plan, warnings, and stop reason. Does not return raw rows or message bodies.","params":["jobId","campaignId","tableId","workspaceId","reportingContext","history"]},{"name":"cancel_campaign_message_preparation","description":"Cancel an active campaign message preparation job and cancel the same pending workflow-table cells that the UI Cancel Pending Cells button would cancel. Launch remains a separate explicit start_campaign step.","params":["jobId","campaignId","tableId","workspaceId"]},{"name":"get_campaign_table_schema","description":"Return compact semantic campaign-table schema: column roles, row count, review-batch metadata, and current approved brief hash. Use for debugging or recovery, not normal queueing.","params":["campaignId","tableId","workspaceId"]},{"name":"select_campaign_cells","description":"Dry-run semantic campaign-cell selection. Returns compact counts and a tiny sample, never bulky row data. For a bounded exact copy audit before approval, call columnRole:\"generateMessage\", rowSelector:{type:\"rowIds\",rowIds:[...]}, includeGeneratedMessageReview:true with 1-20 exact rows. authoritativeGeneratedMessageReview is the WorkflowTableCell.result source of truth for subject, body, and template authority; never substitute formula Message/Subject columns, campaign previews, or CSV exports. For regular refill diagnostics, use rowSelector:{type:\"needsEnrichment\"} to find the earliest unenriched rows by table position, then rowSelector:{type:\"needsApproval\"} to inspect current generated messages that are not approved. needsGeneratedMessage only sees rows that already passed rubric and can skip earlier unenriched rows; use it for generated-message repair, not top-down refill. Pass diagnostic:{limit:N} with 1<=N<=10 and exact workspaceId, campaignId, and tableId to additionally load decision-relevant workflow evidence for that bounded sample in one set query; inert display/formula cells are omitted. Omit diagnostic for the compact default response.","params":["campaignId","tableId","workspaceId","columnRole","rowSelector","limit","includeGeneratedMessageReview","diagnostic"]},{"name":"queue_campaign_cells","description":"Resolve and queue campaign cells by semantic role and bounded row selector. This is the shared campaign-setup, rubric-maintenance, and recovery owner; use it instead of fetching rows only to discover raw cell IDs. For any fill/refill-sends intent, do not call this tool directly: use refill_v3_advance so V3 owns target selection, fencing, and continuation. For approval, rowSelector:{type:\"needsApproval\"} selects current generated rows missing Approved=true, but mutating Approved cells needs explicit bounded approval, or standing customer consent (launchConsent.mode autopilot, or no answer saved; not review_first) for messages that passed the quality check. Do not use needsGeneratedMessage as a top-down cursor because it only selects already-passed rows. forceRerun retries the selected cells only; it never changes the message template or synthesizes messageTemplateRevision authority. A generateMessage rewrite pulls back people whose message was scheduled or ready but not sent (messageRewrite.unscheduledOldCopy); it never touches sent or in-flight messages. After a dos/donts or voice change, use rowSelector passedRows (the brief hash did not change) and repeat with the returned messageRewrite.generatedBefore until queue.queued is 0. People already contacted are never rewritten. For columnRole generateMessage the result carries messageRewrite: the campaign, alreadyContacted (people who keep the message they got), willGetNewMessage, noOneLeft, and the same senders' otherActiveCampaigns. When you change a campaign's message, tell the customer from these facts as create-campaign-v2 references/message-approach.md (Change the message of a running campaign) says; never say Sellable must release, unlock or approve anything. Cross-workspace Agent calls must pass workspaceId.","params":["campaignId","tableId","workspaceId","columnRole","rowSelector","forceRerun","generatedBefore","limit","reason"]},{"name":"wait_for_campaign_processing","description":"Campaign processing wait. Use for fit/message readiness or requireIdle:true when an exact campaign table must settle before replanning. When a preparation job created or owns the frontier, also set requirePreparationTerminal:true; the wait then binds to that durable job (and returns its preparationJobId for exact retries) so zero queued/processing cells cannot falsely settle an undispatched, expired, or still-running cohort. If no job was named and none exists for the campaign, the wait settles on the campaign's cells instead and reports preparationFallback:\"no_preparation_job\". Generated messages are counted separately from rubric passes; current template revision is enforced through messageDraftBasis.approvedBriefHash. With minPassedCount set it also returns ruleConfidence for the campaign rules; rewrite every rule listed in rulesToRewrite before scoring more leads.","params":["campaignId","tableId","workspaceId","minPassedCount","minGeneratedMessages","templateRevision","requireIdle","requirePreparationTerminal","preparationJobId","timeoutMs","intervalMs"]},{"name":"record_campaign_review_batch","description":"Record a fresh campaign-table review batch from explicit row IDs. Use for same-source new-sample recovery after a zero-pass bounded filter run; do not use for new lead sourcing.","params":["tableId","workspaceId","rowIds","enrichCellIds"]},{"name":"revise_message_template_and_rerun","description":"Apply one explicit user-approved message-template revision inside the exact fenced refill run. Requires the complete target config, server run handle, bounded cohort, prior/new authority digests, and request/effect/revision-operation identities. An absent envelope never revises, and forceRerun on another tool cannot synthesize this authority.","params":["campaignId","tableId","workspaceId","targetConfig","runHandle","messageTemplateRevision"]},{"name":"get_auth_status","description":"Verify Sellable authentication early. Call this before any campaign work. Also performs a cached Sellable package update check. Returns ok=false with clear guidance when the token is missing, revoked, or the active workspace is not set. When auth succeeds, use _userNotice when workspace context is useful; it is written in customer-facing product language and may include update guidance. CRITICAL: When auth FAILS with a missing/invalid token (error.type is 'config' or 'auth'), the response includes an `agentInstruction` field: the verbatim FTUX magic-link signup flow for a person, or, for a hosted Sellable agent, what to do while Sellable repairs its access (never ask the customer for an email or login). Follow agentInstruction VERBATIM. Do NOT show the user error.guidance — that's a fallback for non-interactive contexts only.","params":[]},{"name":"get_runtime_identity","description":"Return the exact installed Sellable MCP package identity. Read-only; use it to verify package/runtime provenance after installation.","params":[]},{"name":"report_agent_issue","description":"Log a problem you hit to Sellable's issue log, so the team can fix it. Use it when a run fails, a tool error repeats, or an effect stays uncertain. Never show the customer the code, the summary or a tool name: tell them in plain words what happens next. `code` is a short snake_case code built from plain words (for example refill_run_failed or refill_tool_timeout). `summary` is one plain lowercase sentence about what failed; it must not contain names, emails or URLs. Pass the refs you have (workspaceId, campaignId, tableId, runId). Returns { logged }; it never throws.","params":["code","summary","workspaceId","campaignId","tableId","runId"]},{"name":"get_agent_soul","description":"Read and prove the SOUL.md used by this live Sellable Agent runtime. This tool is runtime-bound and has no target selector.","params":[]},{"name":"update_agent_soul","description":"Replace the SOUL.md used by this live Sellable Agent runtime after a get_agent_soul raw-hash CAS. The current turn keeps its built prompt; the same session rebuilds next turn. If the save outcome is unknown, call get_agent_soul before any retry.","params":["content","expectedRawFileSha256"]},{"name":"start_cli_login","description":"Start the CLI login handoff. Triggers a magic-link email to the supplied email address. Returns the confirm URL the user will be redirected to after clicking the link. Pair with wait_for_cli_login to block until the user finishes browser sign-in.","params":["email"]},{"name":"wait_for_cli_login","description":"Poll the CLI login session every 2s until the user finishes browser sign-in or 5 minutes elapse. On success, writes ~/.sellable/config.json so subsequent MCP calls are authenticated. Re-call with the SAME sessionId on tool_timeout_guard — do NOT call start_cli_login again.","params":["sessionId","timeoutMs"]},{"name":"bootstrap_create_campaign","description":"Run canonical create-campaign preflight in deterministic order: auth, framework, optional resume context, and create-campaign subskill loading.","params":["campaignId","flowVersion","includeTableCheck","refreshContext","host","model","reasoningEffort","modelMetadataSource"]},{"name":"bootstrap_engage","description":"Engage preflight: verify auth/workspace, resolve sender (workspace-first), ensure the workspace's one shared comment campaign table exists, fetch that sender's engaged posts for dedupe, and persist defaults. Rows added later use senderId as assignedSenderId. Minimizes back-and-forth before discovery. IMPORTANT: After auth succeeds, ALWAYS tell the user which workspace is active (from authStatus._userNotice).","params":["senderId"]},{"name":"bootstrap_engage_multi","description":"Multi-sender engage preflight: resolve multiple senders in parallel, ensure one shared comment campaign table for the workspace, fetch engaged posts for all, and return a globalEngagedPostUrls union for cross-sender dedup. The returned per-sender table IDs intentionally converge to that shared table; later add calls still use each senderId to stamp row affinity. Use for multi-sender engage sessions (autonomous or interactive).","params":["senderIds","allSenders"]},{"name":"search_engagement_posts","description":"Search for high-signal LinkedIn posts by keyword for engagement/comment drafting. Designed for the engage skill (no campaignOfferId required). Filters by recency and engagement, can exclude already-engaged URLs, and returns source media hints when the scraper provides images, video covers, or article images.","params":["keywords","page","maxAgeDays","minTotalEngagement","maxPosts","excludePostUrls","targetFollowerMin","targetFollowerMax"]},{"name":"list_workspaces","description":"List accessible workspaces.","params":[]},{"name":"get_active_workspace","description":"Show the active workspace selection.","params":[]},{"name":"get_workspace","description":"Get details of an accessible workspace including members and pending invites.","params":["workspaceId"]},{"name":"set_active_workspace","description":"Set active workspace by ID. Only call this when the user explicitly asks to switch workspace.","params":["workspaceId","userConfirmed"]},{"name":"create_workspace","description":"Create a new workspace. The authenticated user becomes OWNER and the new workspace becomes active for subsequent active-workspace tools.","params":["name"]},{"name":"add_teammate","description":"Invite a teammate to a workspace. Defaults to the active workspace; pass workspaceId to invite to a specific accessible workspace without switching active workspace.","params":["workspaceId","email","role","redirectTo","campaignId"]},{"name":"get_campaign_framework","description":"Load the versioned create-campaign framework (policy, flow, providers, plugins, overrides). Call this at the start of create-campaign.","params":["flowVersion","includePlugins"]},{"name":"bootstrap_find_leads","description":"Create or resume a durable campaignless Find Leads run and its lead list. No campaign or workingCampaignId is required.","params":["bootstrapKey","name"]},{"name":"get_find_leads_run","description":"Read the authoritative durable Find Leads snapshot.","params":["runId"]},{"name":"update_find_leads_run","description":"Persist a provider/source/import approval receipt using optimistic concurrency.","params":["runId","version","action","provider","receipt"]},{"name":"wait_for_find_leads_run","description":"Poll durable Find Leads state until it changes or reaches a terminal state. Progress notifications are optional hints only.","params":["runId","timeoutMs","intervalMs"]},{"name":"cancel_find_leads","description":"Cancel a mutable Find Leads run with its current version.","params":["runId","version"]},{"name":"reissue_find_leads_watch_link","description":"Rotate and return a fresh one-time live Find Leads watch handoff. Old links are revoked and cannot be reconstructed.","params":["runId","mode"]},{"name":"preflight_find_leads_provider","description":"Persist the selected provider prompt/version receipt and return actionable setup diagnostics before any provider call.","params":["runId","runVersion","provider","workflowVersion","promptHash"]},{"name":"get_campaign_context","description":"Return cached create-campaign context (framework snapshot + navigation diagnostics). Refreshes automatically when state is marked dirty.","params":["campaignId","flowVersion","refresh","includeTableCheck"]},{"name":"capture_post_idea","description":"Capture a raw LinkedIn post idea into ~/.sellable/content/linkedin/ideas. Preserves the raw source exactly and returns an ID for later drafting.","params":["rawSource","title","distilledBrief","ideaId","sourceType","sourceUrl","capturedAt"]},{"name":"list_post_ideas","description":"List captured LinkedIn post ideas with compact sanitized previews. Use get_post_idea for full Markdown.","params":["limit"]},{"name":"get_post_idea","description":"Read one captured LinkedIn post idea by ID, including its full Markdown and exact raw source section.","params":["ideaId"]},{"name":"save_hook_research","description":"Save hook research for a LinkedIn post idea under ~/.sellable/content/linkedin/research/hooks.","params":["researchId","ideaId","topic","keywords","sourcePosts","selectedPatterns","previewBudget","notes","createdAt"]},{"name":"save_post_draft","description":"Save a validated LinkedIn post draft under ~/.sellable/content/linkedin/drafts. Use versioned draft IDs for multiple drafts of the same idea.","params":["draftId","ideaId","hookResearchId","priorDraftId","iteration","title","body","validationReceipt","createdAt","status"]},{"name":"calculate_linkedin_hook_preview","description":"Functionally calculate the LinkedIn pre-see-more hook for mobile and desktop. Returns rendered visible lines and rule diagnostics without writing files.","params":["postText","draftId","sourceLabel","measurementMode","requireBrowserMeasurement","clampLines","mobileClampLines","desktopClampLines","mobileTextWidthPx","desktopTextWidthPx"]},{"name":"render_linkedin_post_preview","description":"Render any LinkedIn post text through the Sellable LinkedIn preview CSS contract for mobile and desktop. Writes HTML artifacts and, when local Chrome is available, browser screenshots under ~/.sellable/content/linkedin/previews.","params":["postText","draftId","renderId","sourceLabel","createdAt","renderScreenshots","requireScreenshots","reviewClampLines"]},{"name":"update_post_draft","description":"Update an existing LinkedIn post draft in place while preserving omitted fields. Useful for adding iteration receipts, status changes, or final copy edits without rewriting the whole artifact.","params":["draftId","hookResearchId","priorDraftId","iteration","title","body","validationReceipt","status","updatedAt"]},{"name":"list_post_drafts","description":"List saved LinkedIn post drafts with compact sanitized previews. Use get_post_draft for full Markdown.","params":["limit"]},{"name":"list_post_draft_iterations","description":"List all draft versions for one idea, including iteration scores, verdicts, prior draft links, status, and sanitized previews.","params":["ideaId"]},{"name":"get_post_draft","description":"Read one saved LinkedIn post draft by ID.","params":["draftId"]},{"name":"mark_post_published","description":"Record a published LinkedIn post under ~/.sellable/content/linkedin/published/{year}. Drafts and published posts remain separate files.","params":["draftId","publishUrl","activityId","publishedAt","finalText","title","updateDraftStatus"]},{"name":"get_published_post","description":"Read one published LinkedIn post record by ID. If year is omitted, searches all published year folders.","params":["publishedPostId","year"]},{"name":"update_published_post_metrics","description":"Append a metrics snapshot to a published LinkedIn post record and update latest metric fields for post-performance learning.","params":["publishedPostId","year","capturedAt","metrics","note"]},{"name":"list_published_posts","description":"List published LinkedIn post records with compact sanitized previews.","params":["limit"]},{"name":"get_campaign_navigation_state","description":"Internal/debug: compute deterministic create-campaign navigation state and diagnostics from the saved campaign plus a lightweight table row-count check.","params":["campaignId","includeTableCheck"]},{"name":"get_provider_prompt","description":"Get the search instructions for a lead source provider and register provider prompt preflight readiness. Must be called before provider search/import tools.","params":["provider","campaignOfferId","findLeadsRunId","confirmed"]},{"name":"search_apollo","description":"Search Apollo for people/contacts. Requires get_provider_prompt({ provider: \"apollo\" }) first. Supports two modes: 'people' (default) searches contacts directly, 'combined' (RECOMMENDED for B2B) first finds companies then contacts at those companies. Auto-detects combined mode when company filters are provided.","params":["searchMode","organization_num_employees_ranges","organization_locations","organization_not_locations","q_organization_keyword_tags","organization_industry_tag_ids","latest_funding_amount_min","latest_funding_amount_max","organization_revenue_min","organization_revenue_max","person_titles","person_seniorities","person_locations","include_similar_titles","q_keywords","filters","companyFilters","contactFilters","page","searchId","campaignOfferId","findLeadsRunId","runVersion","targetLeadCount","searchName","companySearchMode","uploadedDomains","currentStep","confirmed"]},{"name":"lookup_sales_nav_filter","description":"Look up filter IDs for Sales Navigator dynamic filters. MUST be called before using REGION, INDUSTRY, CURRENT_TITLE, CURRENT_COMPANY, SCHOOL, or TECHNOLOGIES filters. Returns matching options with IDs that can be used in search_sales_nav.","params":["filterType","query"]},{"name":"search_sales_nav","description":"Search LinkedIn Sales Navigator. Requires exact provider prompt preflight. Pass exactly one owner: campaignOfferId for campaign work or findLeadsRunId plus runVersion for a campaignless run.","params":["filters","keywords","page","searchId","campaignOfferId","findLeadsRunId","runVersion","targetLeadCount","searchName","currentStep","confirmed"]},{"name":"load_csv_domains","description":"Preview and confirm a CSV file that lives on disk, then create a Prospeo domainFilterId for account targeting. This is the CSV-on-disk path for known-account targeting. First call previews the file and returns a confirmationToken. Confirmed execution requires the preview-issued confirmationToken and then returns a domainFilterId you can pass into search_prospeo.","params":["filePath","campaignOfferId","domainColumn","selectedColumns","confirmationToken","confirmed"]},{"name":"load_csv_linkedin_leads","description":"Preview and confirm a CSV file that lives on disk, then create or append to a lead list using LinkedIn profile URLs. This is the person-CSV path. First call previews the file and returns a confirmationToken. Confirmed execution requires the preview-issued confirmationToken, uploads valid LinkedIn profile rows into a lead list, and does NOT use the 1000-domain Prospeo cap.","params":["filePath","campaignOfferId","currentStep","leadListId","leadListName","linkedInColumn","selectedColumns","confirmationToken","confirmed"]},{"name":"list_dnc_entries","description":"Show the active workspace's current Sellable DNC count, list names, and a compact page of entries through the workspace-scoped DNC endpoint. Use this when the user asks to show/check the current DNC list, count, or first page before importing. This is Sellable DNC used by DNC Check, not provider source-list routing.","params":["page","limit","search","listName","sourceType","includeDeleted"]},{"name":"load_csv_dnc_entries","description":"Preview and confirm pasted text, pasted CSV, or a CSV file of domains and LinkedIn profile URLs, then add the valid entries to the active workspace's Sellable DNC list. This is the blocklist/suppression-list path. It reuses Sellable's workspace-scoped DNC endpoints and the campaign DNC Check column, not provider search filters or source-list workarounds.","params":["filePath","csvText","rawText","name","domainColumn","linkedInColumn","confirmationToken","confirmed","workspaceId"]},{"name":"save_domain_filters","description":"Save pasted or raw include/exclude domain lists for Prospeo search. Validates domains and returns a domainFilterId to pass into search_prospeo. Use load_csv_domains when the user already has a CSV file on disk. If the user provides company names, resolve names to official website domains first, then call this tool.","params":["include","exclude"]},{"name":"search_prospeo_companies","description":"Search Prospeo for company/account results through the public /search-company lane. Requires get_provider_prompt({ provider: \"prospeo\" }) first. Use this first for company lookalike/account asks such as \"find companies like Red Rover that use AI and have an API\", \"find founders at companies like our best customers\", or \"find accounts in the news about funding or partnerships\". Results are accounts, not finished people leads. Review the returned accounts with the user, then call confirm_prospeo_company_accounts with the companySearchToken to create a domainFilterId before using search_prospeo for people at those accounts. Supports company lookalike, confirmed AI Attributes including pricing, company news, awards, website search, products/services, integrations, key customers, operating languages, Google discovery, headcount by location, structured company ICP, and experimental website traffic/key executive filters. Do not invent company_oids; when seedCompanies or seedDomains are present, omit company_oids and let the backend resolve real Prospeo IDs. company_intent and support-channel guesses like phone/email/chat/ticket/social are unsupported. Public API rows do not include lookalike tier/score/reason. If output includes omittedFilters or warnings, report them as truthful MCP/backend safety handling rather than claiming Prospeo enforced those omitted constraints. Route lookalike seeds by campaign intent: outbound uses explicit target/account/customer domains or verified past-customer accounts, not the sender company/domain; job-search/application campaigns may use current/past employers as existing-company seeds. Preserve geography like Germany in account or people filters. Company account workflow: search_prospeo_companies -> confirm_prospeo_company_accounts -> search_prospeo. Example: find companies like Red Rover that use AI and have an API:\nsearch_prospeo_companies({\n  \"seedDomains\": [\"tryredrover.com\"],\n  \"filters\": {\n    \"company_lookalike\": { \"minimum_tier\": \"T1\" },\n    \"company_attributes\": { \"uses_ai\": true, \"has_api\": true, \"pricing\": true }\n  }\n})\ncompany_website_search needs a Prospeo plan that includes website search; on PLAN_REQUIRED drop it and keep the other filters. company_technology values must match Prospeo's technology list exactly; on INVALID_FILTERS for a technology, drop that technology rather than guessing spellings.\nDo not invent company_oids. When seedCompanies or seedDomains are present, omit company_oids and let the backend resolve real Prospeo company IDs.\nRoute lookalike seed selection by campaign intent. For outbound/sales prospecting, use explicit target/account/customer domains or company names first, then verified past-customer/account evidence from research, CRM, or proof; never substitute the sender's company/domain or employer history unless it is confirmed as a target/customer seed or the user asks for sender-company peers. For job-search/application campaigns, current/past employers can be lookalike seeds ","params":["campaignOfferId","seedCompanies","seedDomains","filters","limit","page","sort","previewOnly","currentStep","confirmed"]},{"name":"confirm_prospeo_company_accounts","description":"Convert approved Prospeo company account results into a domainFilterId for a follow-on search_prospeo people search. Requires the signed companySearchToken returned by search_prospeo_companies and selectedCompanyIds from that token. Do not submit raw account rows or raw domains as Prospeo-sourced provenance. This does not create people leads directly; it creates the generated-account equivalent of save_domain_filters and returns the next recommended search_prospeo call.","params":["campaignOfferId","companySearchToken","selectedCompanyIds","name","currentStep","confirmed"]},{"name":"search_prospeo","description":"Search Prospeo for people using filters and optional domainFilterId. Requires get_provider_prompt({ provider: \"prospeo\" }) first. Pass exactly one owner: campaignOfferId for campaign work or findLeadsRunId plus runVersion for a campaignless run. Use Prospeo first for hiring-led targeting because it supports company_job_posting_hiring_for and company_job_posting_quantity; Sales Nav does not filter companies by hiring role. When targeting known accounts, call load_csv_domains for CSV-on-disk workflows or save_domain_filters for pasted/raw domain lists, then pass domainFilterId. For current LinkedIn job-post account-source asks, call search_harvest_jobs first, review the artifact, then confirm_harvest_job_companies to create a domainFilterId before using search_prospeo for people. For company lookalike/account asks, call search_prospeo_companies first, review accounts, then confirm_prospeo_company_accounts to create a domainFilterId before using search_prospeo for people. Raw domain inputs, LinkedIn company URLs as domains, and company-name targeting are NOT supported in this MCP tool. Strategy: start with 2-3 high-signal filters (title/seniority + industry or domainFilterId + headcount), add job-posting filters for hiring-led campaigns, then tighten one filter at a time. For security, AppSec, SOC, RevOps, Demand Gen, and similar function-specific lanes, do not widen with bare seniority labels like \"Head\" or \"Director\" alone; pair them with explicit function-title keywords and inspect the sample for off-function `Head of X` leakage. If output includes warnings/fallback, report that the MCP retried a rejected precise title request with a safer domain-filter people search instead of claiming the original precise filter succeeded. Prefer person location over company HQ unless HQ is explicitly needed. Campaign runs MUST pass campaignOfferId so the search shows up in that campaign's Contact Search panel; campaignless Find Leads runs MUST pass findLeadsRunId and runVersion so results stay attached to the durable list. Post-mint create-campaign-v2 watch runs MUST pass currentStep: \"prospeo\". Returns normalized results with pagination.","params":["filters","page","searchId","campaignOfferId","findLeadsRunId","runVersion","targetLeadCount","workspaceId","searchName","domainFilterId","currentStep","confirmed"]},{"name":"search_signals","description":"Search LinkedIn posts for signal discovery. Requires get_provider_prompt({ provider: \"signal-discovery\" }) first. Supports keyword search, profile posts, company posts, or a single post URL. Use campaignless preview mode for directional discovery; include `campaignOfferId` when the search should persist into campaign state. Post-mint create-campaign-v2 watch runs MUST pass campaignOfferId and currentStep: \"signal-discovery\" only after the visible source-plan gate has been approved and leadSourceProvider has been persisted. Omitting campaignOfferId after that approval orphans the search from the UI. Returns a compact summary with recommended posts to avoid context bloat.","params":["type","keywords","profileUrl","postUrl","companyUrl","campaignOfferId","findLeadsRunId","runVersion","currentStep","currentStepTransition","page","limit","headlineICPCriteria","rubricGuidelines","confirmed"]},{"name":"import_leads","description":"Start an approval-gated provider import into exactly one target: campaignOfferId or findLeadsRunId. Campaignless imports materialize directly into the reserved run/list and never read campaign state.","params":["campaignOfferId","findLeadsRunId","runVersion","workspaceId","provider","currentStep","sourceLeadListId","searchId","signalTabId","targetLeadCount","mode","searchName","leadListName","headlineICPCriteria","targetEngagerCount","maxPostsToScrape","allowInvalidSignalPosts","rubricGuidelines","confirmed"]},{"name":"cancel_lead_import","description":"Cancel an in-flight Apollo, Prospeo, or Sales Nav lead import. Use when the operator realizes the current search/filter is wrong and wants to start over without waiting for the broad import to finish. Partial rows already inserted into the lead list are preserved. Provider credits are NOT refunded. After cancellation, call import_leads again with new filters. Signal Discovery cancellation is not supported.","params":["campaignOfferId","workspaceId","tableId","provider"]},{"name":"confirm_lead_list","description":"After the user confirms the lead list looks good, copy the confirmed source list into the campaign table and mark the initial campaign-table execution slice for the flow. This tool owns moving the watched campaign to filter-choice with campaign-setup narration only after non-empty campaign rows exist; do not call update_campaign to fix filter-choice afterward. selectedLeadListId remains the source lead list; workflowTableId is the campaign table.","params":["campaignOfferId","findLeadsRunId","signalTabId","runVersion","workspaceId","sourceLeadListId","currentStep","campaignName","keepInSync","jobId","targetLeadCount","reviewBatchLimit","sourceRowIds","sourceRowLimit","allowPartialSourceList","allowDifferentSourceLeadList","confirmed","includeRawImportResult"]},{"name":"select_promising_posts","description":"Select/promote LinkedIn posts for lead scraping AND provide headline ICP criteria. selectionMode is required: use \"replace\" for a fresh/absolute promoted set that clears previous promoted posts; use \"add\" only when intentionally expanding the current promoted set. Use the selectionTarget returned by search_signals (default 3) for the sampling/promoted set. scrapePlanMode controls the import recommendation: use \"all-selected\" when approval should scrape every promoted post; use \"capacity-target\" with targetEngagerCount when sample math should recommend the smallest right-content post subset that covers the source-candidate target. If selectedCount and recommendedPostCount differ, user-facing copy must describe the recommended/importable scrape set, not the broader selected sample.","params":["campaignOfferId","selections","headlineICPCriteria","selectionMode","scrapePlanMode","targetEngagerCount","maxPostsToScrape","currentStep"]},{"name":"set_headline_icp_criteria","description":"Set 3-5 headline ICP criteria for filtering post engagers. Use when user selects posts manually or attempts to scrape without criteria.","params":["campaignOfferId","headlineICPCriteria","currentStep","workspaceId"]},{"name":"search_harvest_jobs","description":"Search current LinkedIn jobs through Sellable's Harvest-backed v3 API, write markdown/CSV review artifacts, and return a token reference for selected-company confirmation. Use when account-source intent is current LinkedIn job posts. Does not call Prospeo or confirm companies.","params":["search","searches","location","geoId","postedLimit","sortBy","workplaceType","employmentType","experienceLevel","easyApply","under10Applicants","salary","pages","maxRows","artifactFormat","outputDir","fileBaseName"]},{"name":"confirm_harvest_job_companies","description":"Confirm selected Harvest LinkedIn job IDs into a backend-resolved Prospeo domainFilterId, then guide the follow-on search_prospeo people search. Requires searchToken or token reference from search_harvest_jobs. Does not accept raw domains.","params":["searchToken","selectedJobIds","name","outputDir","fileBaseName"]},{"name":"search_inbox_threads","description":"Search LinkedIn inbox threads using the existing Sellable product inbox route. Returns bounded compact summaries for the active workspace, newest activity first. Pass filter \"all\" to see every thread (the route defaults to threads with a draft). Pass messageLimit to get each thread's last N messages inline, so reviewing a window of conversations needs no get_inbox_thread call per thread, and since to drop older threads (reachedSince true means stop paging).","params":["limit","cursor","filter","messageLimit","since","status","senderId","classification","search","includeEnrichment","includeReplyEligibility"]},{"name":"get_inbox_thread","description":"Fetch one LinkedIn inbox thread using the existing Sellable product route GET /api/v3/inbox/[threadId]. Returns bounded thread detail plus a recent chronological transcript for MCP review.","params":["threadId","messageLimit"]},{"name":"check_inbox_reply_eligibility","description":"Load one inbox thread with includeReplyEligibility=true through GET /api/v3/inbox/[threadId] and return the product-computed reply eligibility snapshot.","params":["threadId"]},{"name":"update_inbox_draft","description":"Edit an existing inbox draft through PATCH /api/v3/inbox/[threadId]/draft, then re-fetch the thread and return an approval-required persisted draft snapshot. This does not create a new draft.","params":["threadId","versionId","body"]},{"name":"send_inbox_draft","description":"Send the currently approved existing inbox draft through POST /api/v3/inbox/[threadId]/send after validating the exact approved body hash, draft version, idempotency key, and thread snapshot. No inline body is sent.","params":["threadId","versionId","approvedBody","approvedBodyHash","approvedThreadSnapshot","idempotencyKey","approval","workspaceId"]},{"name":"send_inbox_manual_reply","description":"Send one exact operator-approved manual reply through POST /api/v3/inbox/[threadId]/reply after validating the message hash, idempotency key, and thread snapshot. Batch sends are not supported.","params":["threadId","message","approvedMessageHash","approvedThreadSnapshot","idempotencyKey","approval","workspaceId"]},{"name":"integrations_list_tools","description":"List the provider actions this Agent is allowed to run against a connected app, for one connected account. Returns only reviewed, policy-allowed actions plus a COUNT of the ones that are not allowed, so you can truthfully say there are actions you cannot run without naming them. Call this before integrations_describe_tool or integrations_call_tool. A 'catalog_incomplete' outcome means the provider catalog could not be read completely — it does NOT mean the app has no usable actions. Your own Slack channels are not an integration: read them with read_slack_history.","params":["appSlug","bindingId","accountLabel"]},{"name":"integrations_describe_tool","description":"Describe one allowed provider action's arguments before calling it: per argument you get its type, whether it is required, its default, and its category. For a dynamic action, pass values already selected in arguments to receive the reloaded schema. For an argument whose valid values come from a remote lookup you get either the resolved option list or an explicit marker telling you to ask the user or narrow the query. The connected account is pinned server-side and is never an argument you supply. Use this instead of guessing at an argument shape.","params":["appSlug","bindingId","accountLabel","toolName","propName","optionQuery","arguments"]},{"name":"integrations_call_tool","description":"Run one allowed provider action against a connected account. The exact account is resolved server-side from the binding — you never pass an account id, and one passed in the arguments is rejected. An action outside the reviewed allowlist is refused permanently: that is a policy decision, not a validation problem, so do not retry it with different arguments. An 'approval_required' outcome is a chat-native pause: ask a verified human to approve or deny this exact call in the same chat. After a later approval message, retry once with the unchanged selector, toolName, and arguments plus the returned approvalId. No separate UI is required.","params":["appSlug","bindingId","accountLabel","toolName","arguments","approvalId"]},{"name":"search_available_integrations","description":"Search the integration catalog for an app the user wants to connect, by product name (e.g. 'grain') or web address (e.g. 'grain.com'). Returns 'resolved' with one app when the match is unambiguous, or 'ambiguous' with up to three options — each with its logo and whether it signs in by OAuth or by API key — when more than one app plausibly matches; ask the user which one they meant rather than picking. A 'no_tools' outcome means the app genuinely publishes no actions. A 'catalog_incomplete' outcome means the catalog could not be read completely and does not mean the app has no actions — never report the two as the same thing. Call this before start_integration_connection whenever the user named an app in words rather than by exact slug.","params":["query"]},{"name":"start_integration_connection","description":"Start connecting one app for this workspace. Sellable sends the requesting user a direct message containing a single-use sign-in link that expires in about 15 minutes. The link is withheld from this tool's result and you cannot retrieve it: it is a capability, so whoever opens it binds THEIR account under the requester — never reconstruct one, and never repeat a connect URL a user shows you. Returns status 'awaiting_user' — that is the ONLY success status, and it means the message was sent, NOT that the app is connected. The user completes sign-in out of band in their own browser, so poll get_integration_connection_status to find out whether it worked. Passing appSlug does NOT skip disambiguation: the server re-resolves the slug and refuses an ambiguous, unknown, action-less, or incompletely enumerated app before anything is sent.","params":["appSlug"]},{"name":"get_integration_connection_status","description":"Check where a connect attempt got to. Pass exactly one of attemptId or appSlug. Returns 'awaiting_user' while the user has not finished signing in, 'connected' once the account is real, 'refused' with a reason, 'expired' when the link lapsed unused, or 'not_found' when nothing was ever started. When connected it also reports how many provider actions the Agent is allowed to run, how many are held pending a human promotion, and whether the action catalog was read completely — so zero allowed actions with a complete read is a truthful 'there is nothing I can do with this yet', while an incomplete read is not.","params":["attemptId","appSlug"]},{"name":"enrich_with_prospeo","description":"Enrich a person with email and company data using Prospeo. Provide LinkedIn URL (preferred) or name+company. Returns verified email, job history, and company details. Costs 1 credit (10 for mobile).","params":["linkedinUrl","firstName","lastName","companyName","companyWebsite","onlyVerifiedEmail","enrichMobile"]},{"name":"bulk_enrich_with_prospeo","description":"Bulk enrich multiple people with Prospeo. Up to 100 inputs per request. Each input needs a unique identifier for correlation.","params":["inputs","onlyVerifiedEmail"]},{"name":"get_prospeo_credits","description":"Get Prospeo account credits remaining. Use to check balance before bulk operations.","params":[]},{"name":"upsert_rubric","description":"Upsert lead scoring rubrics for a campaign offer. Pass the full rubric array. Optionally enable ICP filters or delete rubric IDs.","params":["campaignOfferId","rubric","currentStep","enableICPFilters","deletedRubricIds"]},{"name":"draft_rubrics","description":"Generate initial lead scoring rubrics and store them for this campaign. Use save_rubrics to persist.","params":["campaignOfferId","leadScoringRubrics"]},{"name":"select_necessary_rubrics","description":"Keep only the required rubric criteria and mark them as required. Use after draft_rubrics.","params":["campaignOfferId","requiredCheckNames"]},{"name":"save_rubrics","description":"Persist rubric criteria to the campaign. Pass leadScoringRubrics directly to save without drafting. Saving active rubrics enables ICP filtering and keeps the watched client on Filter Rules for user approval; it does not move to Filter Leads or run filtering by itself.","params":["campaignOfferId","leadScoringRubrics"]},{"name":"add_rubric_item","description":"Add a single rubric criterion to the current draft. Call save_rubrics to persist.","params":["campaignOfferId","rubric"]},{"name":"update_rubric_item","description":"Update a single rubric criterion by ID and persist immediately.","params":["campaignOfferId","rubricId","updates"]},{"name":"delete_rubric_item","description":"Delete a single rubric criterion by checkName and persist immediately.","params":["campaignOfferId","checkName"]},{"name":"check_rubric","description":"Legacy compatibility/recovery alias for a bounded saved-rubric sample. It delegates to queue_campaign_cells with the semantic Enrich Prospect role and never discovers raw cell IDs. Do not use it for fill/refill intent; use refill_v3_advance. Normal campaign setup and maintenance should call queue_campaign_cells directly. Cross-workspace Agent calls must pass workspaceId.","params":["campaignOfferId","maxProspects","workspaceId"]},{"name":"wait_for_rubric_results","description":"Wait for ICP rubric pass-rate results to complete for queued prospects. On timeout, returns partial diagnostic stats; create-campaign-v2 should not poll indefinitely. With campaignOfferId it also returns ruleConfidence: per rule, how often the scorer was unsure on the scored sample. Rewrite every rule listed in rulesToRewrite before scoring more leads.","params":["campaignOfferId","tableId","targetCount","minPassedCount","minMessagesCount","timeoutMs","intervalMs","includeRows"]},{"name":"wait_for_campaign_table_ready","description":"Poll backend state until the campaign workflow table exists and (by default) has rows. Use after confirm_lead_list before advancing steps.","params":["campaignId","timeoutMs","intervalMs","requireRows","flowVersion"]},{"name":"wait_for_lead_list_ready","description":"Poll the lead list table until the import completes (default) and rows appear. Use right after import_leads before asking the user to confirm the list.","params":["leadListId","campaignOfferId","findLeadsRunId","workspaceId","provider","jobId","targetLeadCount","timeoutMs","intervalMs","requireRows","requireComplete"]},{"name":"refresh_sender_engagement","description":"Refresh one sender-owned Post Engagers campaign through the product-native CampaignTrackedPost path. Dry-run first returns a human-reviewable receipt; apply requires that dry-run fingerprint and never sends, schedules, approves, prepares messages, launches campaigns, or mutates shared Signal Discovery lanes.","params":["workspaceId","mode","senderId","campaignId","tableId","maxPosts","maxEngagerPages","dryRunFingerprint"]},{"name":"get_table_rows","description":"List leads from a campaign table. Returns: id, name, company, title, linkedinUrl, message (preview), subject (preview), messageCellId, subjectCellId, carryData (extra CSV columns like 'Jobs Hiring For'). Use for selecting which lead to craft.","params":["tableId","limit","page"]},{"name":"get_rows_minimal","description":"List leads from a campaign table with minimal payload. Returns: id, name, company, title, linkedinUrl, carryData (extra CSV columns). Use for fast handoff or enrichment.","params":["tableId","limit","page"]},{"name":"get_rows","description":"Get row details for crafting. Returns: name, company, title, linkedinUrl, currentMessage, currentSubject, messageCellId, subjectCellId, approveCellId, exampleCellId, isExample, carryData (extra CSV columns like 'Jobs Hiring For'), enrichment (summary, experience, education). Use after selecting a lead. messageBeingRewritten true means currentMessage is the old message and a new one is still being written: never quote or judge it as the new message. Do not treat approveCellId from get_rows as the product-visible approval source of truth; on campaign tables it can differ from the send-review Approved column. For route-proof or send-review approval, use get_campaign_messages_preview or select_campaign_cells to select the semantic Approved cell.","params":["tableId","rowIds"]},{"name":"update_cell","description":"Update a cell's value. Use for saving a message (messageCellId/subjectCellId), marking examples (exampleCellId with value: true/false), or approving an explicitly selected semantic Approved cell. Never save prospect copy you wrote from memory: a saved message is the customer's own words or the output of a Sellable message prompt read in full in this context (generate-messages for a campaign template, draft-message for one message). During campaign setup the Generate Message cells write every message (queue_campaign_cells with columnRole generateMessage), so do not write them here. For route-proof or send-review approval, use the approve cell returned by get_campaign_messages_preview or select_campaign_cells, not get_rows alone.","params":["cellId","value","workspaceId"]},{"name":"queue_cells","description":"Legacy exact-cell compatibility/recovery tool. Prefer queue_campaign_cells for campaign work because it resolves semantic column roles and bounded row selectors. For fill/refill intent, use refill_v3_advance instead of reconstructing the refill loop. If exact IDs are required for recovery, pass only the bounded cells already returned by a product owner; do not queue full-table or downstream ICP cells. Cross-workspace Agent calls must pass workspaceId.","params":["tableId","cellIds","workspaceId"]},{"name":"list_subskill_prompts","description":"List available Sellable subskill prompts with visibility controls. Use this when the host does not expose prompts/list.","params":["limit","includePublic","includeInternal"]},{"name":"get_subskill_prompt","description":"Load a Sellable subskill prompt by name. Large prompts are returned in portable chunks by default; continue with nextOffset until hasMore is false. Internal subskills are not listed in prompts/list; use search_subskill_prompts when you are unsure of the exact name.","params":["subskillName","offset","limit"]},{"name":"get_subskill_asset","description":"Load a packaged file that belongs to a Sellable subskill, such as core/flow.v2.json, core/auto-execute.yaml, or references/*.md. Use this instead of reading local repo files so Claude Code and Codex packaged runs share the same config.","params":["subskillName","assetPath","offset","limit"]},{"name":"search_subskill_prompts","description":"Search Sellable subskills by name/description. Use this before get_subskill_prompt if you are unsure which prompt to load.","params":["query","limit","includePublic","includeInternal"]},{"name":"get_source_scout_registry","description":"Return the canonical Sellable source-scout agent names and host filenames for explicit source-comparison/debug runs only. Normal create-campaign source work stays inline in the parent thread.","params":[]},{"name":"get_post_find_leads_scout_registry","description":"Return the canonical Sellable Message Drafting worker name after Start Import copies confirmed source rows. Normal create-campaign uses only this one post-import background agent.","params":[]},{"name":"complete_sender_research","description":"Mark sender research as completed for create-campaign preflight. Call this after research-sender synthesis and before create_campaign.","params":["depth","proofItemsFound","caseStudyItemsFound","credibilitySignalsFound","notes"]},{"name":"fetch_linkedin_posts","description":"Fetch recent LinkedIn posts for a user profile. Returns compact serialized posts with: text, date, reactions, comments, url, isRepost.","params":["linkedinUrl","limit"]},{"name":"fetch_post_engagers","description":"Fetch people who engaged with a LinkedIn post (reactions and/or comments). Returns deduplicated engagers with name, headline, profileUrl, and source. Use 'comments' for lead magnets to get people who actively engaged. Uses smart rate limiting and parallel fetching. In campaign-attached Signal Discovery sampling, call select_promising_posts first to promote the sampled posts into the watched UI.","params":["postUrl","sources","limit"]},{"name":"fetch_linkedin_profile","description":"Fetch LinkedIn profile details. Defaults to compact context (fetchMinimal=true) that strips images/media/provider metadata and returns identity, headline, about, current role, company, limited experience, education, skills, certifications, languages, and projects. Pass fetchMinimal=false only when you need raw provider fields or media URLs. Pass full=true ONLY when you need the long-tail experience history (positions 6+) or complete skill list (skills 3+) for deep research; full still returns compact output unless fetchMinimal=false.","params":["linkedinUrl","full","fetchMinimal"]},{"name":"fetch_company","description":"Fetch LinkedIn company profile details.","params":["companyUrl"]},{"name":"fetch_company_posts","description":"Fetch recent LinkedIn posts from a company page. Returns compact serialized posts with: text, date, reactions, comments, url, isRepost. Use this when asked to find posts by a company.","params":["companyUrl","limit","sortBy"]},{"name":"get_user_posts","description":"Fetch recent LinkedIn posts. Returns compact serialized posts with: text, date, reactions, comments, url, isRepost.","params":["linkedin_url","max_posts"]},{"name":"get_linkedin_profile","description":"(Deprecated) Use fetch_linkedin_profile instead. Fetch LinkedIn profile details.","params":["linkedin_url"]},{"name":"create_on_demand_table","description":"Create an on-demand campaign table backed by a lightweight CampaignOffer. Returns tableId + campaignOfferId.\n\nINPUTS: Pass EITHER clientProspectId (existing prospect ID) OR senderLinkedinUrl (sender's LinkedIn profile URL). One is required.","params":["name","senderIds","clientProspectId","senderLinkedinUrl","offerPositioning","campaignBrief"]},{"name":"create_on_demand_campaign","description":"Create a lightweight general CampaignOffer + workflow table, set senders, and initialize sequence columns. This is not the caller-authored direct InMail path: when exact recipients, Subject, and Message are already supplied, use add_to_inmail_campaign and start_direct_campaign instead. Auto-selects template by sender tiers unless provided.\n\nINPUTS: Pass EITHER clientProspectId (existing prospect ID) OR senderLinkedinUrl (sender's LinkedIn profile URL). One is required.","params":["name","clientProspectId","senderLinkedinUrl","senderIds","campaignBrief","offerPositioning","sequenceTemplate","confirmed"]},{"name":"add_on_demand_leads","description":"Add leads to an on-demand campaign table by LinkedIn URL (dedupes by default).","params":["tableId","leads","skipDuplicates"]},{"name":"start_on_demand_campaign","description":"Start a general on-demand campaign table (set campaignStatus ACTIVE). Do not use this for a typed direct InMail/Connection table; use start_direct_campaign with the matching type.","params":["tableId"]},{"name":"pause_on_demand_campaign","description":"Pause an on-demand campaign table (set campaignStatus PAUSED).","params":["tableId"]},{"name":"init_on_demand_sequence","description":"Initialize sequence columns for an on-demand table. Uses a default template if none provided.","params":["tableId","template","confirmed"]},{"name":"opportunity_sequence_next_page","description":"Read one bounded page of server-selected Lightfield sequence candidates, due work, and exact inbound transitions. Finish a returned continuation before starting a new checkpoint scan.","params":["continuation","checkpoint","limit"]},{"name":"opportunity_sequence_enroll","description":"Idempotently enroll exactly one server-issued Lightfield candidate. The token binds opportunity, contact, trigger, playbook, and source facts; no caller selector is accepted.","params":["candidateToken"]},{"name":"opportunity_sequence_record","description":"Validate one bounded Hermes output against a server-issued work token, reconcile the provider artifact/event identity, append structured Lightfield history first, and then repair the Opportunity projection.","params":["workToken","output"]},{"name":"add_to_inmail_campaign","description":"Add leads with caller-authored Subject and Message copy to a sender's direct InMail execution queue. This is the canonical one-off InMail path when recipients and exact copy are already known; it auto-creates the correctly typed table and dedupes by default. Do not create a normal campaign or a Generate Message column for this shape.","params":["senderId","leads","skipDuplicates"]},{"name":"add_to_connection_campaign","description":"Add leads with caller-authored Message copy to a sender's direct Connection execution queue. This is the canonical one-off connection-plus-DM path when recipients and exact copy are already known; it auto-creates the correctly typed table and dedupes by default.","params":["senderId","leads","skipDuplicates"]},{"name":"add_to_comment_campaign","description":"Add posts to the workspace's one shared Comment direct campaign table. senderId stamps assignedSenderId on each action row so execution uses the intended sender, and duplicate checks remain sender-scoped. Auto-creates the shared table if missing.","params":["senderId","posts","skipDuplicates"]},{"name":"add_to_post_campaign","description":"Add original LinkedIn posts to the workspace's one shared Post direct campaign table. senderId stamps assignedSenderId on each action row so it publishes from the intended sender, and text dedupe remains sender-scoped. Auto-creates the shared table if missing. Rows publish only after Approved is set; each assigned sender's Daily Post Limit (default 1, max 3) spaces approved posts across the sending window. Text-only (max 3000 characters).","params":["senderId","posts","skipDuplicates"]},{"name":"start_direct_campaign","description":"Start the canonical direct campaign after its rows/copy are reviewed and the user explicitly approves launch (sets status to ACTIVE). comment_campaign and post_campaign are workspace-shared, so starting either enables reviewed rows for every assigned sender in that shared table; inmail_campaign and connection_campaign remain sender-specific. Use the same type passed to the add tool. Do not use start_campaign for a table created by add_to_inmail_campaign or add_to_connection_campaign.","params":["senderId","type"]},{"name":"pause_direct_campaign","description":"Pause a canonical direct campaign (sets status to PAUSED). comment_campaign and post_campaign are workspace-shared, so pausing either affects rows for every assigned sender in that shared table; inmail_campaign and connection_campaign remain sender-specific.","params":["senderId","type"]},{"name":"get_or_create_direct_campaign_table","description":"Inspect or explicitly create the canonical direct campaign workflow table for a type. comment_campaign and post_campaign are workspace-shared; senderId adds the sender to the shared table and is later stamped as assignedSenderId when rows are added. inmail_campaign and connection_campaign remain sender-specific. Usually call the matching add_to_* tool instead because it creates the table and rows together. Returns the tableId and whether it was created.","params":["senderId","type"]},{"name":"get_engaged_posts","description":"Get all post URLs a sender has already engaged with or has scheduled. In the workspace-shared Comment Campaign, history is filtered by each row's assignedSenderId so dedupe remains sender-scoped.","params":["senderId"]},{"name":"get_company_info","description":"Return the active workspace's canonical Company Info markdown document plus updatedAt metadata. Read this before calling set_company_info so you can pass expectedUpdatedAt and avoid overwriting someone else's edits.","params":[]},{"name":"set_company_info","description":"Replace the active workspace Company Info markdown document in full through the narrow Company Info API. Use real newlines, never literal \"\\n\" sequences. Pass expectedUpdatedAt from get_company_info for normal writes; only pass forceOverwrite: true when the user deliberately approves overwriting the latest version. Omit companyInfoMarkdown or pass null/empty/whitespace to clear Company Info.","params":["companyInfoMarkdown","approvalId","expectedUpdatedAt","forceOverwrite"]},{"name":"get_billing","description":"Read the active workspace's billing: plan (free, solo, team, legacy_seat, managed or none), status, whether sending is allowed and why not, trial and plan end dates, LinkedIn accounts in use, credits (balance, included per month, next top-up, low, steps waiting for credits, prices) and the exact links to start or resume a plan, buy credits, upgrade to Team, add a LinkedIn account or manage the card and invoices. Call it for any question about the plan, trial, credits, prices, payment, invoices or why sending stopped, and answer from it with its exact links. Read-only; it changes nothing.","params":[]},{"name":"customer_program","description":"Observe the home workspace campaign relationship stage, read prior complete daily or weekly campaign results, and retain campaign context. Scheduled stage_delivery grants native posting only when delivery.deliverNow is true; false means stay silent and reconcile. refill_summary claims the one delivery of a closed refill run's customer summary (trigger scheduled, runId); stage its text with stage_delivery and askIdentity. When the customer says no to an offer (a create_connection_campaign ask), call decline_ask with askIdentity set to that ask's identity; the offer is not made again for 30 days for those senders. A resume_sends ask is the pause check-in: senders whose campaigns a person paused and who sent nothing for a whole sending day; keep paused calls decline_ask and it is asked again after 7 days, and start or new campaign calls accept_ask. A reconnect or an out-of-people ask cannot be declined. When the customer says yes to an ask, call accept_ask with that ask's identity, then do the work; it is not asked again for 30 days, unless a yes to start a campaign started one that ran out too: the next offer then comes back in sendCoverage.asks. A start option's notContacted is the people that campaign can still contact; the campaigns an ask lists have no one left, so never quote their notContacted. claim_review, daily_results and weekly_review return sendCoverage: whether every sender is topped off for its next sending day, schedule (what is scheduled for each day refill looked ahead to: written lines that say for each sender what is ready out of what the day holds and the reason for anything not full (a day with more than three senders says who is full in one line, then only the exceptions), and fixes, what only the customer can do to fill a sender), and asks ranked by unfilled capacity (biggest first). Tell the customer the schedule day blocks as written (schedule.tell), never what was added; the refill_summary claim carries the same schedule; while sendCoverage.asks is not empty a review may not record maintain or refill. sendCoverage.agentWork is your own next step (more people for a running campaign always; a connection campaign or a restart under standing consent): do it, then record continue with the operation id, or propose_create for a connection campaign. sendCoverage.acceptedOpen lists asks the customer already answered yes: say it is Working. Each ask carries since, daysOpen and missedSoFar: write it as refill-sends references/customer-summary.md (Write an ask) says. When the customer replies always to an out-of-people ask, save_context launchConsent {mode: autopilot, decidedAt: now}, then start the first start option with pausedByCustomer false; under standing consent (launchConsent autopilot, or no answer saved; only review_first is not), start such an option yourself whenever one is offered and say so instead of asking. Never start a campaign the customer paused (pausedByCustomer true) unless they ask. decisionRef and operationRef ","params":["action","context","preferences","trigger","leaseOwner","effectId","fence","reviewAction","summary","decisionRef","operationRef","text","includesDailyResults","runId","askIdentity"]},{"name":"read_slack_history","description":"Read your own Slack channels: recent messages (each with its reply count) or every reply in one thread. Use it whenever someone asks you to look back over the channel or read another thread, for example what issues keep coming up over the past two weeks. It reads only the channels you were added to, through your own Slack app, so never search integrations or offer to connect Slack for this. Without channel it reads your home channel; days sets how far back (1-30, default 14). To read a thread, pass its ts as threadTs. Messages come oldest first; when hasMore is true call again with before set to nextBefore (channel) or after set to nextAfter (thread). When Slack says slack_rate_limited, wait retryAfterSeconds and repeat the same call. For conversations you took part in, session_search (when you have it) is quicker. Only a Sellable Agent can call this tool.","params":["channel","days","threadTs","before","after","limit"]},{"name":"get_sender_routing","description":"Return the workspace's territory routing rules markdown plus the connected sender roster. If configured is false, routing is inactive and all campaign senders are eligible everywhere.","params":["workspaceId"]},{"name":"set_sender_routing","description":"Store free-form sender territory routing markdown verbatim for the active workspace. Use real newlines, never literal \"\\n\" sequences. Geography, segments, verticals, and named-account exceptions are all valid. Omit rulesMarkdown or pass an empty string to clear the rules and disable routing. Senders not mentioned in rules remain globally eligible. The response echoes the stored markdown and full sender roster; verify every sender is covered by a rule or intentionally unrestricted, then confirm coverage with the user.","params":["rulesMarkdown","workspaceId"]},{"name":"list_senders","description":"List outbound sender identities available in the active workspace, including LinkedIn profile URL and verified identity proof when available. In create-campaign flows, use this at Settings/final launch handoff to verify available connected senders; auto-select only when a sender has an exact identity match to the researched campaign identity. If identity proof is missing, ask the user to confirm the sender.","params":["workspaceId"]},{"name":"get_sender","description":"Get a specific outbound sender identity by senderId (workspace-scoped), including generationContext with that sender's saved stories, voice, source limits, company information and explicit saved/default/unbound mode. This read does not select a campaign: read the selected campaign separately and prefer its scoped offer over the labeled company fallback. Missing or unbound personal context is not another sender's biography. No settings are changed.","params":["senderId","workspaceId"]},{"name":"refresh_paid_inmail_credits","description":"Refresh the cached paid InMail credit facts for one exact sender in the explicit request-scoped workspace by calling Sellable's scoped MCP credit refresh path. This updates only the sender credit cache/attempt metadata (balance, sentSince, checkedAt, last attempted/error/lease) and returns the sanitized coordinator receipt. It does not mutate campaigns, lower thresholds, approve messages, schedule sends, start campaigns, send messages, spend InMail credits, or write scheduler fields. refill_sends --yolo normally handles get_refill_target_plan refresh_paid_inmail_credits candidates internally; use this standalone tool for manual diagnostics or explicit one-off refreshes, then rerun get_refill_target_plan before any refill mutation.","params":["senderId","workspaceId","actionType","campaignId","tableId","columnId","threshold","maxStalenessSeconds","paidInmailCreditsMaxStalenessSeconds"]},{"name":"set_sender_calendar_link","description":"Set or clear one exact outbound sender's calendar booking link through Sellable's workspace-scoped compare-before-write route. Read list_senders first and pass its calendarLink and calendarLinkRevision. Use force only after the user explicitly authorizes overwriting newer state.","params":["senderId","workspaceId","calendarLink","expectedCalendarLink","expectedCalendarLinkRevision","force"]},{"name":"set_sender_context","description":"Save one exact outbound sender's personal context: the four lists the message writer uses as that sender's fact bank and voice (workStories, personalNotes, voiceExamples, sourcesAndLimits). This is what a sender interview produces. It replaces the sender's whole saved context; it cannot touch another sender, the calendar link, writing style, dos/donts or Company Info. Read get_sender first and pass generationContext.updatedAt as expectedUpdatedAt (null when nothing is saved yet). Limits: at most 10 entries per list, 3000 characters per entry, 15000 UTF-8 bytes in all. Only write what the sender said or confirmed; never copy one sender's facts into another's. Pass senderContext: null to clear it.","params":["senderId","workspaceId","senderContext","expectedUpdatedAt"]},{"name":"get_engage_state","description":"Get persisted engage defaults (per active workspace) to avoid relying on LLM memory. Includes lastSenderId, lastSenderIds (multi-sender), lastCommentCampaignTableIds, and lastMode.","params":[]},{"name":"set_engage_state","description":"Persist engage defaults (per active workspace) so future runs can auto-select sender/table without relying on LLM memory. Supports both single-sender (lastSenderId) and multi-sender (lastSenderIds) fields.","params":["lastSenderId","lastCommentCampaignTableId","clearCommentCampaignTableId","lastSenderIds","lastCommentCampaignTableIds","lastMode"]},{"name":"get_engage_memory","description":"Load unified Sellable memory from ~/.sellable/configs/: core identity/company/proof/story memory, transcript/content-memory clusters, references, style and post writing rules, proven search keywords, and tracked people. The get_engage_memory name is backward-compatible; all data lives in readable markdown files the user can edit directly. When senderId is provided, reads compatibility overrides from senders/{senderId}/ with flat-path fallback.","params":["senderId"]},{"name":"set_engage_style_guide","description":"Write the user's commenting style guide. When senderId is provided, writes to ~/.sellable/configs/senders/{senderId}/styleguide-core.md. Otherwise writes to ~/.sellable/configs/writing/styleguide-core.md.","params":["markdown","senderId"]},{"name":"record_engage_proven_search","description":"Record a keyword search outcome. When senderId is provided, writes to senders/{senderId}/proven-searches.md. Otherwise writes to discovery/proven-searches.md. Updates running averages if the keyword already exists.","params":["keyword","totalReturned","postsYielded","senderId"]},{"name":"upsert_engage_tracked_person","description":"Add or update a person in tracked people. When senderId is provided, writes to senders/{senderId}/influencers.md. Otherwise writes to discovery/influencers.md.","params":["name","linkedinUrl","reason","senderId"]},{"name":"copy_sender_config","description":"Copy all per-sender config files (styleguide-core.md, proven-searches.md, influencers.md) from one sender to another. Useful for bootstrapping a new sender from an existing one.","params":["fromSenderId","toSenderId"]},{"name":"migrate_flat_configs","description":"One-time migration: copy existing flat configs (writing/styleguide-core.md, discovery/proven-searches.md, discovery/influencers.md) into senders/{senderId}/. Leaves originals in place. Run once per sender when transitioning to per-sender config.","params":["senderId"]},{"name":"create_workflow_table","description":"Create a campaign-backed workflow table with a lightweight CampaignOffer shell and minimal config columns optionally derived from sequence action types. Returns tableId, campaignOfferId, column list, and derived requirements.\n\nAll tables include LinkedIn URL, Name, and Approved. Additional columns depend on the planned actions: INVITE/DM -> Message, INMAIL_* -> Subject + Message, COMMENT -> Post URL + Comment + Reaction Type. Campaign brief, rubrics, positioning, and prospect linkage remain optional follow-on metadata.\n\nINPUTS: Pass EITHER clientProspectId OR senderLinkedinUrl. Both optional — if neither is supplied, the table is created without prospect linkage.","params":["name","senderId","clientProspectId","senderLinkedinUrl","sequenceActions"]},{"name":"attach_sequence","description":"Attach a sequence template to an existing workflow table. This works for direct campaign tables and for type-agnostic tables created with create_workflow_table. The tool pre-validates template structure and surfaces 409 conflicts clearly when sequence columns already exist.","params":["tableId","template","templateRef","confirmed","currentStep"]},{"name":"attach_recommended_sequence","description":"Attach the tier-recommended sequence template to a campaign. Delegates template selection to the backend's existing tier-aware selector (selectTemplateForTiers), which picks the unified Sales Nav cascade for Sales Nav/Recruiter senders: Open Profile check -> Open InMail, otherwise fresh paid credits >= 5 -> paid InMail, otherwise same-campaign connection request -> accepted wait -> DM. Basic, Premium, or mixed sender sets still get `INVITE -> accepted -> DM`. This is the default Sales Nav product path; use attach_sequence directly only when the caller needs a custom non-recommended cadence. Prefer this over attach_sequence during the create-campaign-v2 autonomous tail — it removes the risk of hand-authoring an invalid template mid-long-context.","params":["campaignId","confirmed","currentStep"]},{"name":"list_tables","description":"List all workflow tables in the active workspace. Returns metadata including workspaceId, workflow status, campaignStatus, dashboardBucket, whether each table is campaign-backed, and whether it has a sequence attached.\n\nUnlike get_campaigns, this lists every WorkflowTable regardless of how it was created, including create_workflow_table, create_on_demand_table, and UI-created tables.","params":["limit","hasSequence","kind","webhookEnabled"]},{"name":"create_table","description":"Create an empty standalone WorkflowTable in an explicitly named workspace, without creating a campaign, sender, sequence, or lead-list surrogate. Optionally add registered columns and one server-credentialed inbound webhook. The webhook URL is sensitive and is returned only in this authorized private creator result; never copy it into logs, evidence, screenshots, shared documents, or list output. Agent-service calls require requester confirmation; after approval, retry the exact same arguments with the returned approvalId. An idempotencyKey makes exact retries converge, but the key is optional, so the tool is truthfully not globally idempotent.","params":["workspaceId","approvalId","idempotencyKey","name","description","columns","inboundWebhook"]},{"name":"export_table_csv","description":"Export a workflow table using the existing Sellable Export CSV endpoint. Optional filters/sort are passed through and splitRows writes deterministic CSV batches.","params":["tableId","filters","sort","splitRows"]},{"name":"export_workspace_csv","description":"Export active-workspace outreach CSVs to local files on the MCP host. Use exportIntent to disambiguate counts before exporting: `reached_out_people` means one deduped row per unique prospect who received an actual contact attempt (invite, DM, open InMail, or paid/closed InMail) and is the right choice for 'everyone we reached out to'; `dashboard_messages` means message events only (DMs + InMails), matching the dashboard Messages Sent card; `connections_sent` means invite events only; `outreach_event_log` means one row per action event for audit/reconciliation; `people_and_events` exports the reached-out people CSV plus matching contact event rows. If the user asks for 'sent', 'reached out', or cites conflicting dashboard/export numbers, ask which intent they want before calling the tool. Requires a valid Sellable API key plus active workspace; run list_workspaces and set_active_workspace first if needed. The manifest includes file paths, counts, filters, package/runtime metadata, and skipped/deferred scope notes. CSV files may contain prospect and outreach metadata, so store them appropriately.","params":["exportType","exportIntent","datasets","actionTypes","outputDir","fromSentAt","toSentAt","campaignIds","tableIds"]},{"name":"list_column_types","description":"List workflow-table column options MCP can discover. No filters besides includeHidden/includeDeprecated are supported; visible createable options are returned by default.","params":["includeHidden","includeDeprecated"]},{"name":"get_column_schema","description":"Get MCP discovery metadata for one workflow-table column option, including editor contract, mutation capabilities, editable fields, full replacement config mode, runCells behavior, and hidden/deprecated safety verdicts. Prefer optionId when a columnType has multiple visible presets.","params":["optionId","columnType"]},{"name":"commit_blueprint","description":"Validate and commit a full workflow-table blueprint (multiple columns + wiring) in one call. Call get_column_schema first for createable config fields and runCells behavior. Use when you have designed every column up-front. NOT for incremental additions — use add_column for that.","params":["tableId","blueprint","validateOnly","runCells"]},{"name":"add_column","description":"Validate and add a single column to an existing workflow table. Call get_column_schema first for createable config fields, default config, and runCells behavior. Use validateOnly:true to dry-run. Use http_request named \"Enrich Prospect\" for enrichment, score_icp_mcp for ICP scoring, and http_request named \"Generate Message\" with /api/v1/outbound-builder/generate-message for Sellable message generation. The native generate_message type is retired and load-only. NOT for batch additions — use commit_blueprint.","params":["tableId","column","validateOnly","runCells"]},{"name":"update_column","description":"Update a workflow-table column name or full replacement config. Call get_column_schema first for mutation capabilities and declared editable fields. Backend read-only/capability errors are returned as structured, model-fixable results. Use runCells:true only when schema says runCellsOnEdit is supported or an explicit backend no-op is acceptable.","params":["tableId","columnId","name","config","runCells"]},{"name":"delete_column","description":"Delete a workflow-table column through the backend cleanup route. Call get_column_schema first for mutationCapabilities.delete; backend read-only/capability errors pass through as structured, model-fixable results.","params":["tableId","columnId"]},{"name":"reorder_columns","description":"Reorder workflow-table columns using exact backend order semantics. Call get_column_schema first for mutationCapabilities.reorder on existing column types, then supply every column ID in desired order. Backend capability errors pass through as structured, model-fixable results.","params":["tableId","orderedColumnIds"]}],"toolsHash":"f6ea6bf3881b6346f72837812b734541d362178ef80ba2958fffee6c2791b5a7","wrote":["/tmp/.sellable/update-check.json"],"callsAt":"2026-11-16T11:32:40.427Z","calls":[{"tool":"list_agent_endpoints","ok":true,"ms":5,"bytes":193},{"tool":"get_campaigns","ok":false,"ms":2,"bytes":0,"error":"tool reported an error: Error: Sellable not configured.\n\nRun a Sellable workflow to start first-run login, for example /sellable:create-campaign in Claude Code or $sellable:create-campaign in Codex.\n\nIf the browser login pag"},{"tool":"get_refill_target_plan","ok":false,"ms":0,"bytes":0,"error":"tool reported an error: Error: Sellable not configured.\n\nRun a Sellable workflow to start first-run login, for example /sellable:create-campaign in Claude Code or $sellable:create-campaign in Codex.\n\nIf the browser login pag"},{"tool":"get_campaign_table_schema","ok":false,"ms":1,"bytes":0,"error":"tool reported an error: Error: Sellable not configured.\n\nRun a Sellable workflow to start first-run login, for example /sellable:create-campaign in Claude Code or $sellable:create-campaign in Codex.\n\nIf the browser login pag"},{"tool":"get_runtime_identity","ok":true,"ms":1,"bytes":126}],"host":{"commit":"eaaffffaf52255d1b77dc4aacffdfaf353f0c12c","run":"https://github.com/itsHaddad/bragproof/actions/runs/43857070216"},"network":{"names":["registry.npmjs.org"],"conns":[],"callNames":[],"callConns":[]}},"summary":{"outcome":"started","needs":[],"contacts":[],"flags":[{"kind":"very-long","tool":"refill_sends","quote":"3000 characters or more"},{"kind":"very-long","tool":"setup_evergreen_campaigns","quote":"3000 characters or more"},{"kind":"orders","tool":"get_auth_status","quote":"er-facing product language and may include update guidance. CRITICAL: When auth FAILS with a missing/invalid token (error.type is 'config' or 'auth'), the response incl"},{"kind":"orders","tool":"bootstrap_engage","quote":"ssignedSenderId. Minimizes back-and-forth before discovery. IMPORTANT: After auth succeeds, ALWAYS tell the user which workspace is active (from authStatus._userNotice)."},{"kind":"orders","tool":"lookup_sales_nav_filter","quote":"Look up filter IDs for Sales Navigator dynamic filters. MUST be called before using REGION, INDUSTRY, CURRENT_TITLE, CURRENT_COMPANY, SCHOOL, or TECHNOLOGIES fi"},{"kind":"very-long","tool":"search_prospeo_companies","quote":"3000 characters or more"},{"kind":"orders","tool":"search_prospeo","quote":"er company HQ unless HQ is explicitly needed. Campaign runs MUST pass campaignOfferId so the search shows up in that campaign's Contact Search panel; campaignless F"},{"kind":"orders","tool":"search_signals","quote":"nto campaign state. Post-mint create-campaign-v2 watch runs MUST pass campaignOfferId and currentStep: \"signal-discovery\" only after the visible source-plan gate ha"},{"kind":"very-long","tool":"customer_program","quote":"3000 characters or more"}],"tools":200,"wrote":1,"callsTried":5,"callsAnswered":2,"callContacts":[]},"history":[{"package":"@sellable/mcp","version":"0.1.1020-wip.morningrefill.20261113.1","ranAt":"2026-11-16T11:32:29.064Z","outcome":"started","tools":200,"contacts":0,"flags":9,"serious":0,"toolsHash":"f6ea6bf3881b6346f72837812b734541d362178ef80ba2958fffee6c2791b5a7"},{"package":"@sellable/mcp","version":"0.1.1010-wip.scheduledbyday.20261112.3","ranAt":"2026-11-12T17:51:43.997Z","outcome":"started","tools":200,"contacts":0,"flags":9,"serious":0,"toolsHash":"8080e27f514093d93cfa0c68445e8b07b36501d8777a805c19bdf4ac9c2b290a"},{"package":"@sellable/mcp","version":"0.1.1009-wip.noreadback.20261112.1","ranAt":"2026-11-12T16:55:16.416Z","outcome":"started","tools":200,"contacts":0,"flags":9,"serious":0,"toolsHash":"0c053c8979163af1a18add992a3c4e890b6cabfd762333b0a1533065493ad1fd"}],"page":"https://bragproof.smaverk.com/r/%40sellable/mcp"}