@sellable/mcp 0.1.444 → 0.1.445

package/dist/tools/campaign-message-preparation.js

@@ -6,7 +6,7 @@ async function postPrepareMessages(body) {
 export const campaignMessagePreparationToolDefinitions = [
     {
         name: "start_campaign_message_preparation",
-        description: 'Start a bounded message-preparation job for a specific existing CampaignOffer campaignId/tableId. This is the active_campaigns existing-row path after resolve_campaign_fill_route, exact target re-read, and active prep-job check; it is not campaign creation and not evergreen horizon fill. It never launches the campaign, sends messages, or directly writes scheduledFor. The job queues pending Enrich Prospect cells first, lets ICP/rubric and Generate Message cascade, then marks ready or approves only the bounded cohort. 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.',
+        description: 'Start a bounded message-preparation job for a specific existing CampaignOffer campaignId/tableId. This is the active_campaigns existing-row path after resolve_campaign_fill_route, exact target re-read, and active prep-job check; it is not campaign creation and not evergreen horizon fill. It never launches the campaign, sends messages, or directly writes scheduledFor. The job processes 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. Use this or rowSelector:{type:"needsEnrichment"} for regular refill frontiers; 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.',
         inputSchema: {
             type: "object",
             properties: {

package/dist/tools/campaign-processing.d.ts

@@ -5,6 +5,12 @@ type RowSelector = {
 } | {
     type: "rowIds";
     rowIds: string[];
+} | {
+    type: "needsEnrichment";
+    limit?: number;
+} | {
+    type: "needsApproval";
+    limit?: number;
 } | {
     type: "passedRows";
     limit?: number;

package/dist/tools/campaign-processing.js

@@ -40,7 +40,7 @@ export const campaignProcessingToolDefinitions = [
     },
     {
         name: "select_campaign_cells",
-        description: "Dry-run semantic campaign-cell selection. Returns compact counts and a tiny sample, never bulky row data. Use before queue_campaign_cells only when debugging or explaining a recovery action.",
+        description: "Dry-run semantic campaign-cell selection. Returns compact counts and a tiny sample, never bulky row data. 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.",
         inputSchema: {
             type: "object",
             properties: {
@@ -64,6 +64,8 @@ export const campaignProcessingToolDefinitions = [
                             enum: [
                                 "reviewBatch",
                                 "rowIds",
+                                "needsEnrichment",
+                                "needsApproval",
                                 "passedRows",
                                 "needsGeneratedMessage",
                                 "staleGeneratedMessages",
@@ -84,7 +86,7 @@ export const campaignProcessingToolDefinitions = [
     },
     {
         name: "queue_campaign_cells",
-        description: "Resolve and queue campaign cells by semantic role and row selector. Normal create-campaign tail should use this instead of fetching rows only to discover cell IDs. Use forceRerun:true for message-template revisions.",
+        description: "Resolve and queue campaign cells by semantic role and row selector. Normal create-campaign tail should use this instead of fetching rows only to discover cell IDs. For regular refill enrichment, queue columnRole:\"enrich\" with rowSelector:{type:\"needsEnrichment\"} or use start_campaign_message_preparation for the full top-down prep loop. For approval, rowSelector:{type:\"needsApproval\"} selects current generated rows missing Approved=true, but mutating Approved cells still requires explicit bounded approval. Do not use needsGeneratedMessage as a refill cursor because it only selects already-passed rows. Use forceRerun:true for message-template revisions.",
         inputSchema: {
             type: "object",
             properties: {
@@ -102,6 +104,8 @@ export const campaignProcessingToolDefinitions = [
                             enum: [
                                 "reviewBatch",
                                 "rowIds",
+                                "needsEnrichment",
+                                "needsApproval",
                                 "passedRows",
                                 "needsGeneratedMessage",
                                 "staleGeneratedMessages",

package/dist/tools/campaign-refill-state.js

@@ -6,7 +6,7 @@ async function postCampaignRefillState(body) {
 export const campaignRefillStateToolDefinitions = [
     {
         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 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.",
+        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.",
         inputSchema: {
             type: "object",
             properties: {

package/dist/tools/prompts.js

@@ -373,7 +373,7 @@ export function getPostFindLeadsScoutRegistry() {
             codex: 'After confirm_lead_list copies source rows and the initial campaign-table execution slice exists, ask the filter-choice question immediately. Do not spawn anything before that question. After the answer, launch only Message Drafting. The filter-choice answer is the post-import user gate for this single worker; do not ask another question about starting it in step-wise or YOLO mode. The registry lookup is not a launch: after get_post_find_leads_scout_registry, immediately invoke Task/spawn_agent or the host background-agent mechanism before loading filter-leads.md, before saving rubrics, and before treating skip-filters as ready for message review. Both choices route through this kickoff; do not let filters_skipped jump straight from filter-choice to message-generation. If filters are chosen, the parent stays on Filter Rules and drafts/saves rubrics with MCP tools while Message Drafting runs in the background. If filters are skipped, move to Messages/message review only after Message Drafting has started or is ready; update_campaign(currentStep=messages) is not proof of launch. If the named Message Drafting custom agent is unavailable, spawn a generic gpt-5.5 xhigh Message Drafting background agent with the same lean campaign/table basis. When the background worker starts, persist workerDetails.messageDraftBuilder with statusSource "branch", status "branch-running", runId, startedAt, updatedAt, basisToken when known, and basis containing campaignId, selectedLeadListId, workflowTableId, filterChoice, and reviewBatchRowHash or reviewBatchRowIds; workerStatuses.messageDraftBuilder may be "running" as a simple badge only. Never put rich proof under workerStatuses and never use workerStatuses.messageDrafting. If no background-agent tool is callable, start the same full message branch inline before filter drafting or before skip-filter message review, record workerDetails.messageDraftBuilder with statusSource "parent-thread-fallback" and status "fallback-active", and require the same live context, prompt, assets, and validation gate before message review; do not wait until filters are saved and then call the registry.',
             claude: "After confirm_lead_list copies source rows and the initial campaign-table execution slice exists, ask the filter-choice question immediately. Do not invoke any Task/Agent before that question. After the answer, invoke only Message Drafting. If filters are chosen, parent drafts/saves rubrics with MCP tools while Message Drafting runs, asks filter approval, then joins Message Drafting. If filters are skipped, invoke only Message Drafting and move to Messages/message review.",
             parentThreadRule: 'Named agents are optional acceleration, but message drafting is not optional. The only normal background worker is Message Drafting. The filter-choice answer is the campaign-scoped go-ahead for this single post-import worker; do not ask another question to start it in step-wise or YOLO mode. If a named agent is unavailable, use a generic gpt-5.5 xhigh Message Drafting background agent. source work and filter work stay in the parent thread with MCP tools. If post-find-leads-message-scout is available, run it as the background Message Draft Builder after the filter-choice answer. The registry lookup is not a launch: get_post_find_leads_scout_registry only identifies the worker, and Message Drafting counts as started only after Task/spawn_agent or the host background-agent tool is invoked, or after the parent begins the same full message branch inline because no background-agent tool is callable. This launch must happen before loading filter-leads.md, save_rubrics, filter approval, or skip-filter message review; currentStep=messages is not proof of launch. If post-find-leads-message-scout is absent, do not customer-surface install status. Do not silently treat message drafting as started; the main thread must either launch the background worker or execute the same message branch from CampaignOffer state, selected source state, workflowTableId, and initial campaign-table execution slice rows. For a spawned worker, record workerDetails.messageDraftBuilder with statusSource branch / status branch-running, runId, startedAt, updatedAt, and basis containing campaignId, selectedLeadListId, workflowTableId, filterChoice, and reviewBatchRowHash or reviewBatchRowIds. workerStatuses.messageDraftBuilder is optional simple badge text only ("running", "ready", "blocked", "idle"); never put runId/statusSource/basis under workerStatuses and never use workerStatuses.messageDrafting. If no background-agent tool is callable, start that same full message branch inline before filter drafting or before skip-filter message review, record workerDetails.messageDraftBuilder with statusSource parent-thread-fallback / status fallback-active then ready, and require the same live context, prompt, assets, and validation gate before message review; do not report that as a background worker failure. If neither branch nor inline fallback can run, return blocked/retry-needed; do not wait until filters are saved and then call the registry. The Message Drafting handoff must be lean. Do not paste copied row counts, brief hashes, review-batch hashes, full reviewBatchRowIds, broad row data, or local debug artifacts into the spawn prompt. Local markdown/json files are not normal-path inputs. The filter-choice question is the first post-import user gate; do not load post-lead registries or filter references before it. Message drafting starts after the filter-choice answer, must load get_subskill_prompt({ subskillName: "generate-messages" }), and must load every required message asset named by generate-messages Mode 0 through get_subskill_asset before drafting. Reference Asset Loading means loading the required pre-draft reference pack before drafting; return blocked/retry-needed if required assets cannot be loaded; load ai-tells.md because it is never optional. The branch or parent-thread fallback loads the full generate-messages prompt and every referenced asset through get_subskill_asset. After generating/revising the candidate and before returning ready, must load get_subskill_prompt({ subskillName: "create-campaign-v2-validation" }) as the final internal validation gate, must read live campaign table state through scoped MCP/product tools, and must reject mismatched selectedLeadListId/workflowTableId/campaign/workspace input. Do not block when filters were chosen but leadScoringRubrics are not yet visible in the branch read; the parent owns save_rubrics and filter approval in parallel, so Message Drafting should return status ready with basisStatus usable_initial when campaign/list/table identity and the non-empty execution slice match. Do not use any alternate, local-artifact, or examples-only message prompt. User copy feedback, message QA, or rewrite requests before approve-message must be routed back to Message Drafting with the current recommendation, lean campaign/table basis, and latest user text; the parent must not rewrite or QA the template from memory and must not call update_campaign_brief before approve-message. The worker validates internally and returns only templateRecommendation, tokenFillRules, renderedGoodSample, status, approveOrReviseRecommendation, validationStatus, outputAt, outputHash, and blocked/retry detail. Do not render renderedFallbackSample, risk notes, or a qaReceipt on the normal happy path. On the filter path, save_rubrics keeps the browser on Filter Rules after save_rubrics so the user can approve the saved criteria; after saved-filter approval, move to Filter Leads with currentStep=apply-icp-rubric whether Message Drafting is ready or still running. Wait there for message approval. Enrichment, filtering, Generate Message cells, sender setup, sequence attach, and launch wait for template approval on the Use Template path. On the skip path, move to Messages/message review after Message Drafting has started or is ready and wait for message approval before enrichment or Settings. Do not render message review from checklist or shortcut instructions; message review requires a messageDraftRecommendation whose basis proves the generate-messages prompt, required message assets, and validation gate ran for the current campaign/table execution slice. Do not automatically rerun Message Drafting after filters/enrichment finish; show the initial draft by default and offer an enriched rewrite only with explicit user opt-in. Handoff and recommendation output are Markdown with labeled fields, not raw JSON.',
-            prepareMessagesRule: `Default create-campaign stays on the existing reviewBatchLimit:15 first campaign-table execution slice. For plain post-mint fill/load/refill requests, load get_subskill_prompt({ subskillName: "refill-sends-workflow" }), then call resolve_campaign_fill_route({ intent:"plain" }), list_senders, and get_campaign_refill_state for enough candidate campaigns to identify the campaign that most recently had scheduler-owned sends for the relevant sender set before any mutation. Route outcomes are route:"evergreen_horizon", route:"active_campaigns", and route:"ask_create"; if a plain managed-waterfall route has archived/completed skipped slots or does not cover the named sender set, immediately refetch with resolve_campaign_fill_route({ intent:"active" }) and inspect current dashboard-active ACTIVE/PAUSED campaign-backed sequence campaigns before declaring a sender blocked; stay in the same campaignOfferId/campaignId context after minting. Plain fill is not an alias for fill_campaign_horizon or campaign creation. fill_campaign_horizon is evergreen-only and is not the generic regular-campaign fill path. Campaign creation is allowed only after route:"ask_create" and explicit user selection; it is never the default response to plain fill. Treat "fill up/load sends" as capacity-fill preparation, and treat "refill senders", "fill senders", "max out senders", and "load everyone up" as sender-scoped capacity-fill preparation. For sender-scoped requests with no named senders, --yolo means all eligible healthy senders enrolled in active campaign-backed sequence campaigns in the active workspace; without --yolo, ask which eligible enrolled senders to refill before choosing campaigns or mutating. In non-yolo interactive Codex or Claude Code sessions, the sender/campaign/action approval packet must be asked through request_user_input/AskUserQuestion with exactly Accept and Decline; do not use plain chat approval. The approval question body must show workspace, sender scope, a campaign-by-campaign table, exact ids/caps/dates, side effects, forbidden actions, and stop condition before mutation. Default --yolo horizon is a two-send-day horizon: two sender-local send days, skipping no-send-hour days. For campaign-scoped fill/refill, select the best recent-send campaign and calculate the bounded gap from healthy sender daily capacity minus future scheduler-owned scheduled sends and ready-to-schedule rows, then prepare only that gap. For sender-scoped fill/refill, calculate the bounded gap per eligible sender across active enrolled campaigns, counting future scheduled and ready-to-schedule rows across those campaigns, then choose the best same-sender campaign to fill each sender gap: prefer recent/future scheduler-owned sends for that sender, then strongest recent result evidence, then source health. Maintain a horizon saturation ledger per selected sender with selected send days, gross capacity, future scheduler-owned scheduled counts, ready-to-schedule buffer, remaining scheduled gap, remaining ready-or-scheduled gap, and the next MCP primitive. Do not stop after filling only one sender when the request was sender-scoped. In --yolo, continue through every safe selected sender/campaign action covered by the rendered packet, reread after each terminal apply/prep/start result, recompute the horizon saturation ledger, and keep going until scheduler-owned scheduledFor cells fill the requested horizon or a concrete blocker/timeout is proven. Ready-to-schedule rows are buffer, not completion; if ready covers the gap but scheduled cells do not, run a bounded scheduler-settle loop and report awaiting_scheduler_after_ready_buffer as not complete when timeout is reached. When no safe in-packet action remains return concrete continuation options with campaign names, exact ids, and which options require a new approval packet. Use the refill workflow to decide whether to enrich/prep more rows in that same recent-send/best-result campaign, add/import more rows to that same campaign/source path, use a different existing campaign only when the selected same-sender lane is blocked/exhausted/already loaded while the sender still has a gap, or ask what to create. Do not create warm-post-engager side campaigns. Surface sender-health blockers separately from prepared/approved/scheduled counts. User-facing refill decisions must be campaign-name-first and sender-name-first, with ids as proof/execution targets only. For already-running regular campaigns that need Signal Discovery source replenishment, use the guarded currentStep clear with clearCurrentStepIfMatches:"running", campaign-scoped provider prompt/search/select, and import_leads with the existing sourceLeadListId when a newly approved selected-post scrape would otherwise return reusedExistingSourceList. After confirm_lead_list copies rows into an existing table, avoid fixed maxRowsToCheck:100; if confirm_lead_list returns USER_ADDED_ROWS_LIMIT_EXCEEDED, create a bounded same-source split from selectedLeadListId with get_rows_minimal/load_csv_linkedin_leads, confirm that smaller source list into the same campaign, and operate on the reviewBatch first; inspect reviewBatch/table selectors or use adaptive/wider bounded prep so appended rows are included. Do not interpret checkedRows as enriched rows; it is only the table cursor. Prepared, approved, and ready_to_schedule rows are intermediate states; never call them scheduled unless a re-read proves scheduler-owned scheduled cells with non-null scheduledFor fill the requested horizon, and never call them complete unless a final re-read proves the same. Before source import, prep, approval, or selected paused campaign start, require exact visible approval or --yolo packet auto-accept and a fresh get_campaign_refill_state reread; stop if freshness.stateHash or exact ids changed. For "approve X messages", use approvalMode:approve only when explicitly requested. For "schedule X sends" or "fill sender sends", approve only when explicitly requested, then re-read campaign/table scheduled counts; if scheduler-owned scheduledFor cells are not present, report prepared/approved/ready - awaiting scheduler instead of success. Do not call start_campaign as part of fill/schedule horizon. start_campaign is allowed in refill only for exact selected PAUSED, dashboard-active, campaign-backed sequence targets named in the bounded packet, and the packet must state that starting can let the product scheduler schedule/send approved eligible sequence actions. Never start unrelated, archived, completed, draft, direct, or non-selected campaigns, never broad approve-all, and never use direct scheduler writes. campaignId is CampaignOffer.id. If the user asks to stop preparation, the target is wrong, or status shows the wrong campaign/table, use cancel_campaign_message_preparation only for the exact active job. Low-level selectors are diagnostics and recovery only for this lane.`,
+            prepareMessagesRule: `Default create-campaign stays on the existing reviewBatchLimit:15 first campaign-table execution slice. For plain post-mint fill/load/refill requests, load get_subskill_prompt({ subskillName: "refill-sends-workflow" }), then call resolve_campaign_fill_route({ intent:"plain" }), list_senders, and get_campaign_refill_state for enough candidate campaigns to identify the campaign that most recently had scheduler-owned sends for the relevant sender set before any mutation. Route outcomes are route:"evergreen_horizon", route:"active_campaigns", and route:"ask_create"; if a plain managed-waterfall route has archived/completed skipped slots or does not cover the named sender set, immediately refetch with resolve_campaign_fill_route({ intent:"active" }) and inspect current dashboard-active ACTIVE/PAUSED campaign-backed sequence campaigns before declaring a sender blocked; stay in the same campaignOfferId/campaignId context after minting. Plain fill is not an alias for fill_campaign_horizon or campaign creation. fill_campaign_horizon is evergreen-only and is not the generic regular-campaign fill path. Campaign creation is allowed only after route:"ask_create" and explicit user selection; it is never the default response to plain fill. Treat "fill up/load sends" as capacity-fill preparation, and treat "refill senders", "fill senders", "max out senders", and "load everyone up" as sender-scoped capacity-fill preparation. For sender-scoped requests with no named senders, --yolo means all eligible healthy senders enrolled in active campaign-backed sequence campaigns in the active workspace; without --yolo, ask which eligible enrolled senders to refill before choosing campaigns or mutating. In non-yolo interactive Codex or Claude Code sessions, the sender/campaign/action approval packet must be asked through request_user_input/AskUserQuestion with exactly Accept and Decline; do not use plain chat approval. The approval question body must show workspace, sender scope, a campaign-by-campaign table, exact ids/caps/dates, side effects, forbidden actions, and stop condition before mutation. Default --yolo horizon is a two-send-day horizon: two sender-local send days, skipping no-send-hour days. For campaign-scoped fill/refill, select the best recent-send campaign and calculate the bounded gap from healthy sender daily capacity minus future scheduler-owned scheduled sends and ready-to-schedule rows, then prepare only that gap. For sender-scoped fill/refill, calculate the bounded gap per eligible sender across active enrolled campaigns, counting future scheduled and ready-to-schedule rows across those campaigns, then choose the best same-sender campaign to fill each sender gap: prefer recent/future scheduler-owned sends for that sender, then strongest recent result evidence, then source health. Maintain a horizon saturation ledger per selected sender with selected send days, gross capacity, future scheduler-owned scheduled counts, ready-to-schedule buffer, remaining scheduled gap, remaining ready-or-scheduled gap, and the next MCP primitive. Do not stop after filling only one sender when the request was sender-scoped. In --yolo, continue through every safe selected sender/campaign action covered by the rendered packet, reread after each terminal apply/prep/start result, recompute the horizon saturation ledger, and keep going until scheduler-owned scheduledFor cells fill the requested horizon or a concrete blocker/timeout is proven. Ready-to-schedule rows are buffer, not completion; if ready covers the gap but scheduled cells do not, run a bounded scheduler-settle loop and report awaiting_scheduler_after_ready_buffer as not complete when timeout is reached. When no safe in-packet action remains return concrete continuation options with campaign names, exact ids, and which options require a new approval packet. Use the refill workflow to decide whether to enrich/prep more rows in that same recent-send/best-result campaign, add/import more rows to that same campaign/source path, use a different existing campaign only when the selected same-sender lane is blocked/exhausted/already loaded while the sender still has a gap, or ask what to create. Do not create warm-post-engager side campaigns. Surface sender-health blockers separately from prepared/approved/scheduled counts. User-facing refill decisions must be campaign-name-first and sender-name-first, with ids as proof/execution targets only. For already-running regular campaigns that need Signal Discovery source replenishment, use the guarded currentStep clear with clearCurrentStepIfMatches:"running", campaign-scoped provider prompt/search/select, and import_leads with the existing sourceLeadListId when a newly approved selected-post scrape would otherwise return reusedExistingSourceList. Before prep, inspect get_campaign_refill_state.preparationFrontier. If hasLaterPreparedIsland:true or earliestUnpreparedRow exists before later successful enrichment, use rowSelector:{type:"needsEnrichment"} with columnRole:"enrich" in table-position order or start_campaign_message_preparation adaptive defaults; do not use the UI Jump anchor or needsGeneratedMessage as the refill cursor. After confirm_lead_list copies rows into an existing table, avoid fixed maxRowsToCheck:100; if confirm_lead_list returns USER_ADDED_ROWS_LIMIT_EXCEEDED, create a bounded same-source split from selectedLeadListId with get_rows_minimal/load_csv_linkedin_leads, confirm that smaller source list into the same campaign, and inspect reviewBatch only as diagnostics unless the approved packet explicitly prioritizes the just-copied split and earlier needsEnrichment rows are exhausted, dependency-blocked, or excluded. For approval diagnostics use rowSelector:{type:"needsApproval"} after current generated messages exist; mutating Approved cells still requires approvalMode:approve and exact bounded approval. Do not interpret checkedRows as enriched rows; it is only the table cursor. Prepared, approved, and ready_to_schedule rows are intermediate states; never call them scheduled unless a re-read proves scheduler-owned scheduled cells with non-null scheduledFor fill the requested horizon, and never call them complete unless a final re-read proves the same. Before source import, prep, approval, or selected paused campaign start, require exact visible approval or --yolo packet auto-accept and a fresh get_campaign_refill_state reread; stop if freshness.stateHash or exact ids changed. For "approve X messages", use approvalMode:approve only when explicitly requested. For "schedule X sends" or "fill sender sends", approve only when explicitly requested, then re-read campaign/table scheduled counts; if scheduler-owned scheduledFor cells are not present, report prepared/approved/ready - awaiting scheduler instead of success. Do not call start_campaign as part of fill/schedule horizon. start_campaign is allowed in refill only for exact selected PAUSED, dashboard-active, campaign-backed sequence targets named in the bounded packet, and the packet must state that starting can let the product scheduler schedule/send approved eligible sequence actions. Never start unrelated, archived, completed, draft, direct, or non-selected campaigns, never broad approve-all, and never use direct scheduler writes. campaignId is CampaignOffer.id. If the user asks to stop preparation, the target is wrong, or status shows the wrong campaign/table, use cancel_campaign_message_preparation only for the exact active job. Low-level selectors are diagnostics and recovery only for this lane.`,
         },
     };
 }

package/dist/tools/setup-evergreen-campaigns.js

@@ -6,7 +6,7 @@ async function postSetupEvergreenCampaigns(body) {
 export const setupEvergreenCampaignsToolDefinitions = [
     {
         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. 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. 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 acceptedRuntimes, rejectedRuntimes, requiresVisibleThreadOrDurableReceipt, receiptArtifactHint, receiptRunId, and receiptMustBeWrittenAfter; pre-existing receipts at old deterministic paths are stale and must not be used, so stop with blocked:stale_receipt_artifact if the receipt was not freshly written for the current receiptRunId. `multi_agent_v1.spawn_agent`/opaque spawn_agent is not accepted for mutating command proof unless the parent has visible thread or durable receipt proof. In local Codex, prefer `codex_app.list_projects` then `codex_app.create_thread` with a local project target; do not create a worktree for lane execution. If Codex app thread tools are unavailable but local Codex CLI is available, use durable streaming workers with `codex -a never -s danger-full-access -c model_reasoning_effort=xhigh exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> -`; approval, sandbox, and reasoning-effort config flags must appear before `exec`, and current customer CLI installs reject `codex exec --ask-for-approval never` and `codex exec -a never`. Do not rely on default model or default reasoning effort: copy the parent model such as `gpt-5.5` into `-m` and always pass `-c model_reasoning_effort=xhigh`; a child worker that reports GPT 5.5 with high reasoning is a launcher bug to relaunch before mutation, not a user-continue path. Plan responses include approvalSummary; render approvalSummary when asking for bounded delegated approval because it explicitly lists campaignsToCreate, campaignsToUpdate, campaignsToVerifyOnly, campaignsLeftUntouched, attachedSenders, selectedActionIds, allowedSideEffects, forbiddenSideEffects, blockers, and approvalQuestion. When safe-yolo needs normal setup work, the parent skill may ask for bounded delegated approval: one approval over the current planRevision, selected action ids, caps, allowed side-effect classes, and stop conditions lets lane workers execute without per-substep approval while staying inside that packet. In exec/automation mode, do not call request_user_input; if yolo plan autoExecutable:false and no interactive approval can be received, stop with blocked:bounded_approval_unavailable_in_exec_mode before any mutation. Lane workers must explicitly load and use the installed `$sellable:create-campaign` wrapper as the nested workflow entrypoint, then load `create-campaign-v2` and `create-campaign-v2/core/flow.v2.json`; they must execute creation, source import, create-campaign workflow steps, generate-messages, sequence attachment, pause_campaign review-state transition when the current table is still DRAFT, and review readiness through that existing create-campaign workflow/subskills, then return receipts here for verification. Customer-visible verify receipts must set status:'succeeded' or status:'completed'; status:'passed', status:'pass', and status:'passed_with_warnings' are rejected as primary success statuses. Exception: a Post Engagers lane may write status:'blocked' with blocker:'post_engagers_no_sender_posts', 'post_engagers_no_recent_sender_posts', 'post_engagers_no_sender_owned_posts', or 'post_engagers_source_author_mismatch' when Sellable MCP readback proves no usable sender-authored posts/source exists; verify returns these as acceptedLaneBlockers so the parent can report the no-op instead of retrying invalid source repair. Any other blocked receipt, including model-quality/preflight or worker model availability blockers, is not a goal-complete success condition and must not let the parent report evergreen completion. Receipts must include createCampaignStepReceipt with setupPlanCall, createCampaignWorkflowReceipt, campaignBriefReceipt, sourceDecisionReceipt, filterDecisionReceipt, messageDraftingReceipt, reviewBatchReceipt, sequenceReceipt, and verifyCall nested inside createCampaignStepReceipt; top-level-only copies of those objects are not enough and are not promoted by verify. setupPlanCall must use canonical keys: planRevision, actionId, laneKey, workspaceId, senderIds, campaignId, tableId, createIntent. Do not use laneActionId, lanePacketActionId, delegatedPlanRevision, delegatedActionId, or requestedCall text as a substitute for those canonical fields. createCampaignWorkflowReceipt must include skillCommand:'$sellable:create-campaign', skillName:'create-campaign', wrapperSkillLoaded:true, workflowPromptName:'create-campaign-v2', workflowPromptLoadedToHasMoreFalse:true, workflowAssetPath:'create-campaign-v2/core/flow.v2.json', workflowAssetLoaded:true, workerRuntime, workerThreadId or receiptArtifactPath, durableReceiptWritten when using a receipt file, and notAdHoc:true. messageDraftingReceipt must use exactly statusSource:'branch' or statusSource:'packaged-generate-messages-worker'; descriptive aliases such as statusSource:'package-readback-local-thread' are rejected. It must include proof that generate-messages was loaded, start_campaign_message_preparation/get_campaign_message_preparation_status ran when the packaged worker path is used, validationResult:'passed', a passed qualityReview, and at least 3 concrete sampleMessages with rowId, generatedMessageText, verdict, and issues; Do not substitute `message` for `generatedMessageText`; Do not substitute `passVerdict` for `verdict`. Before writing durable receipts, run a receipt self-check: top-level `planRevision`, `actionId`, `laneKey`, `laneType`, `workspaceId`, and `senderIds` must exist; if the self-check fails, fix the receipt before ending. Use start_campaign_message_preparation with approvalMode:\"mark_ready\" only for evergreen setup. Never call `start_campaign_message_preparation` with `approvalMode:\"approve\"`; approve exactly one semantic Approved cell through select_campaign_cells/update_cell and final proof must show approvedGeneratedMessageCount exactly 1. Shared Cold Fallback samples with a standalone name followed by 'Hey there' are rejected. This command does not launch campaigns, does not schedule sends, does not assign scheduler-owned send fields, does not raw-write campaign status, does not archive/delete cleanup targets, and does not spend paid credits.",
+        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. 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. 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 acceptedRuntimes, rejectedRuntimes, requiresVisibleThreadOrDurableReceipt, receiptArtifactHint, receiptRunId, and receiptMustBeWrittenAfter; pre-existing receipts at old deterministic paths are stale and must not be used, so stop with blocked:stale_receipt_artifact if the receipt was not freshly written for the current receiptRunId. `multi_agent_v1.spawn_agent`/opaque spawn_agent is not accepted for mutating command proof unless the parent has visible thread or durable receipt proof. In local Codex, prefer `codex_app.list_projects` then `codex_app.create_thread` with a local project target; do not create a worktree for lane execution. If Codex app thread tools are unavailable but local Codex CLI is available, use durable streaming workers with `codex -a never -s danger-full-access exec -C <repo> -o <worker-final-file> -`; approval and sandbox flags must appear before `exec`, and current customer CLI installs reject `codex exec --ask-for-approval never` and `codex exec -a never`. Plan responses include approvalSummary; render approvalSummary when asking for bounded delegated approval because it explicitly lists campaignsToCreate, campaignsToUpdate, campaignsToVerifyOnly, campaignsLeftUntouched, attachedSenders, selectedActionIds, allowedSideEffects, forbiddenSideEffects, blockers, and approvalQuestion. When safe-yolo needs normal setup work, the parent skill may ask for bounded delegated approval: one approval over the current planRevision, selected action ids, caps, allowed side-effect classes, and stop conditions lets lane workers execute without per-substep approval while staying inside that packet. In exec/automation mode, do not call request_user_input; if yolo plan autoExecutable:false and no interactive approval can be received, stop with blocked:bounded_approval_unavailable_in_exec_mode before any mutation. Lane workers must explicitly load and use the installed `$sellable:create-campaign` wrapper as the nested workflow entrypoint, then load `create-campaign-v2` and `create-campaign-v2/core/flow.v2.json`; they must execute creation, source import, create-campaign workflow steps, generate-messages, sequence attachment, pause_campaign review-state transition when the current table is still DRAFT, and review readiness through that existing create-campaign workflow/subskills, then return receipts here for verification. Customer-visible verify receipts must set status:'succeeded' or status:'completed'; status:'passed', status:'pass', and status:'passed_with_warnings' are rejected as primary success statuses. Exception: a Post Engagers lane may write status:'blocked' with blocker:'post_engagers_no_sender_posts', 'post_engagers_no_recent_sender_posts', 'post_engagers_no_sender_owned_posts', or 'post_engagers_source_author_mismatch' when Sellable MCP readback proves no usable sender-authored posts/source exists; verify returns these as acceptedLaneBlockers so the parent can report the no-op instead of retrying invalid source repair. Receipts must include createCampaignStepReceipt with setupPlanCall, createCampaignWorkflowReceipt, campaignBriefReceipt, sourceDecisionReceipt, filterDecisionReceipt, messageDraftingReceipt, reviewBatchReceipt, sequenceReceipt, and verifyCall nested inside createCampaignStepReceipt; top-level-only copies of those objects are not enough and are not promoted by verify. setupPlanCall must use canonical keys: planRevision, actionId, laneKey, workspaceId, senderIds, campaignId, tableId, createIntent. Do not use laneActionId, lanePacketActionId, delegatedPlanRevision, delegatedActionId, or requestedCall text as a substitute for those canonical fields. createCampaignWorkflowReceipt must include skillCommand:'$sellable:create-campaign', skillName:'create-campaign', wrapperSkillLoaded:true, workflowPromptName:'create-campaign-v2', workflowPromptLoadedToHasMoreFalse:true, workflowAssetPath:'create-campaign-v2/core/flow.v2.json', workflowAssetLoaded:true, workerRuntime, workerThreadId or receiptArtifactPath, durableReceiptWritten when using a receipt file, and notAdHoc:true. messageDraftingReceipt must use exactly statusSource:'branch' or statusSource:'packaged-generate-messages-worker'; descriptive aliases such as statusSource:'package-readback-local-thread' are rejected. It must include proof that generate-messages was loaded, start_campaign_message_preparation/get_campaign_message_preparation_status ran when the packaged worker path is used, validationResult:'passed', a passed qualityReview, and at least 3 concrete sampleMessages with rowId, generatedMessageText, verdict, and issues; Do not substitute `message` for `generatedMessageText`; Do not substitute `passVerdict` for `verdict`. Before writing durable receipts, run a receipt self-check: top-level `planRevision`, `actionId`, `laneKey`, `laneType`, `workspaceId`, and `senderIds` must exist; if the self-check fails, fix the receipt before ending. Use start_campaign_message_preparation with approvalMode:\"mark_ready\" only for evergreen setup. Never call `start_campaign_message_preparation` with `approvalMode:\"approve\"`; approve exactly one semantic Approved cell through select_campaign_cells/update_cell and final proof must show approvedGeneratedMessageCount exactly 1. Shared Cold Fallback samples with a standalone name followed by 'Hey there' are rejected. This command does not launch campaigns, does not schedule sends, does not assign scheduler-owned send fields, does not raw-write campaign status, does not archive/delete cleanup targets, and does not spend paid credits.",
         inputSchema: {
             type: "object",
             properties: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sellable/mcp",
-  "version": "0.1.444",
+  "version": "0.1.445",
   "type": "module",
   "description": "Sellable MCP server for Claude Code and Codex campaign workflows",
   "main": "dist/index.js",

package/skills/create-campaign-v2/references/ai-tells.md

@@ -56,12 +56,10 @@ raise your hand for X (creepy to reach out based on that, i know) - but
 this felt too on the nose to ignore"` shapes only when the source was an
 explicit lead-magnet comment, reply, or opt-in.
 
-Shared lane handling: for Shared Signal Discovery, a high-level topic bridge is
-allowed and usually preferred when source evidence supports the topic:
-`"Saw you around conversations about X, so hope this is relevant."` Keep it at
-topic level and do not recap the source, reaction, comment, post, author, or
-scraper path. For Shared Cold Fallback, source/conversation language is still a
-tell; start from a cold buyer problem or row-supported observation instead.
+Shared evergreen exception: for Shared Signal Discovery, `hope this is
+relevant` and topic-level `saw you in a few conversations about/around X, so
+hope this is relevant` are allowed when evidence-backed and not engagement-
+claiming. `Might be interested` remains blocked because it assumes intent.
 
 **Severity:** REJECT
 

package/skills/create-campaign-v2/references/gold-standard-message-examples.md

@@ -203,9 +203,60 @@ Why it works:
 **Proof ranks used:** rank 1 (mechanism - 1,000+ conditions from one blood
 draw), rank 3 (risk before claims).
 
-Use the Superpower weak-signal hedge pattern as a valid model for Shared Signal Discovery when the source topic is real, third-party, and high-level enough to mention without activity-log phrasing. Shared Cold Fallback must not use this source language.
+### Dotwork Shared Signal Discovery Gold Standard
 
-### Shared Evergreen Signal Discovery Topic Bridge Pattern
+Use this as the gold standard when a shared Signal Discovery lane has real
+topic evidence, an InMail format, and enough founder/operator proof to make the
+note feel earned.
+
+```text
+Subject: initiatives + business outcomes + one live picture
+
+Hey {{first_name}},
+
+Saw you might be interested in {{signal_topic}}, so hope this is relevant.
+
+I helped build {{credible_prior_company}}, so I've seen firsthand where {{buyer_category}} tools fall short.
+
+{{product}} keeps {{buyer_object_1}}, {{buyer_object_2}}, and {{buyer_object_3}} connected in one live model.
+
+It also gives AI agents structured context to work from, instead of stale exports.
+
+Open to a quick look at how {{company}}'s operating model would map in?
+
+P.S. We're backed by {{proof_point}} as a modern alternative to {{competitor_set}}.
+```
+
+Why it works:
+
+- the subject follows the plus-sign A + B + C pattern: topic + outcome + concrete picture
+- the opener is topic-level and low-certainty without exposing scrape mechanics
+- the credibility line explains why the sender has earned the point of view
+- the product is specific, not three generic product-summary sentences
+- the CTA offers a concrete mapped view for the recipient's company
+- the PS uses proof to answer a category objection, not to brag
+
+Transfer limits:
+
+- Use `Saw you might be interested in [topic], so hope this is relevant` or
+  `Saw you in a few conversations about/around [topic], so hope this is
+  relevant` only for Shared Signal Discovery when the source evidence supports
+  the topic and the line does not claim a reaction, comment, sender-authored
+  post, or other unproven engagement.
+- Do not use `found you in a thread`; that exposes sourcing mechanics.
+- Do not use founder/operator proof unless the brief supports it and every
+  attached sender can credibly send the line.
+- For InMail, keep the body as one cohesive message and use a concrete
+  plus-sign subject like `initiatives + business outcomes + one live picture`.
+
+The Superpower weak-signal hedge can also work for Shared Evergreen Signal
+Discovery when the line is topic-level and evidence-backed. Prefer the Dotwork
+shape above when the brief supports an InMail subject, sender credibility,
+specific product mechanism, company-specific CTA, and PS proof. Use the safe
+pattern below when a concrete signal/problem bridge is stronger than a
+low-certainty source opener.
+
+### Shared Evergreen Signal Discovery Safe Pattern
 
 Use when the campaign is a shared evergreen signal lane and the source signal is
 real but not sender-owned.
@@ -213,22 +264,20 @@ real but not sender-owned.
 ```text
 Hey {{first_name}},
 
-saw you around conversations about agentic AI governance, so hope this is relevant.
-
-Traditional governance often treats AI like predictable software, which leaves a gap between strategy and real-time delivery.
+The AI-assisted GTM conversations keep coming back to one problem: turning good LinkedIn signal into review-ready pipeline without another outbound dashboard.
 
-Persona is a multi-tenant control plane for auditing agentic identities across complex environments.
+Sellable helps teams find signal-matched prospects, filter for fit, and generate reviewed LinkedIn copy from one workflow.
 
-Open to a 15-minute walkthrough?
+Curious if LinkedIn outbound is a channel you're trying to make more reliable this quarter?
 ```
 
 Why it works:
 
-- low-confidence relevance is aimed at source/topic fit, not buyer intent
+- no blocked low-confidence relevance hedge
 - no `I'm building` or founder-only first person
 - no internal workflow vocabulary such as Codex, MCP, agent workflows, workflow
   table, lead source, or signal discovery
-- the signal becomes a buyer problem without a generic LinkedIn-pipeline claim
+- the signal becomes a buyer problem instead of an activity-log opener
 - safe for multiple senders attached to the same shared lane
 
 ## Conditional Examples

package/skills/create-campaign-v2/references/gold-standard-message-patterns.md

@@ -199,24 +199,28 @@ Shape:
 Prefer:
 
 - lowercase or casual casing when the motion supports it
-- an honest line like "hope this is relevant" or "so may be off, but this seemed relevant" for non-evergreen weak-signal campaigns; for Shared Signal Discovery, use the high-level topic bridge override below
+- an honest line like "hope this is relevant" or "so may be off, but this seemed relevant" for Shared Signal Discovery or non-evergreen weak-signal campaigns
 - a concrete CTA that names the useful conversation or asset
 - binary options only when the brief explicitly supports two real next steps
 - direct wording over polished marketing language
 
 Shared Evergreen Signal Discovery override:
 
-- Shared Signal Discovery may use `saw you around conversations about [topic],
-  so hope this is relevant` when source evidence supports a high-level topic.
-- Keep the topic bridge at business-theme level; do not recap the source,
-  reaction, comment, post, author, or scraper path.
-- Shared Cold Fallback must not use `hope this is relevant`, `might be
-  interested`, or `saw you in a few conversations` as the opener.
+- `Saw you in a few conversations about/around [topic], so hope this is
+  relevant` is allowed when the source evidence supports the topic and the line
+  does not claim a reaction, comment, sender-authored post, or other unproven
+  engagement.
+- The Dotwork opener `Saw you might be interested in [topic], so hope this is
+  relevant` is also allowed for Shared Signal Discovery when the source
+  evidence supports the topic. Do not use bare `might be interested` as a
+  generic intent claim.
+- Do not use `found you in a thread`; that exposes sourcing mechanics.
 - Do not use `I'm building`, `I've mapped`, `my company`, or `my team`.
-- Turn the source topic into a buyer problem bridge only when it is more
-  specific and less presumptive than the topic bridge.
-- If the source topic cannot support a high-level bridge, omit the source line and
-  start from a role/company/problem observation.
+- Prefer the Dotwork shape when the message is InMail: a concrete A + B + C
+  subject, topic-level relevance opener, proof of earned sender perspective,
+  specific product mechanism, company-specific CTA, and optional PS proof.
+- If the source topic cannot support a low-certainty topic bridge, omit the
+  source line and start from a role/company/problem observation.
 
 Avoid:
 

package/skills/create-campaign-v2/references/sellable-cleanup-rules.md

@@ -74,7 +74,33 @@ Revise or reject the sample when any of these happen.
 - **actions are implied, not stated** — e.g. "runs that chain as AI agents" when a clearer version would name the specific actions (verb + object, one per line)
 - **category-level opener used when a per-lead signal exists** — if `lead-sample.json` carries any per-lead signal (post, hire, visible tool, topic engagement), the opener must reference it. Category-level openers of shape `"Most [category] teams still do X by hand"` are only acceptable when zero per-lead signal is in the sample. When a category-level opener is used as fallback, Findings must flag it explicitly
 - **mind-reading from engagement signals** — a topic engagement, post, public activity, role, company, or hiring trigger does not prove buyer intent. Reject phrases like `"AI-GTM stack is clearly on your mind"`, `"you're clearly focused on..."`, `"obviously relevant"`, or `"already thinking about..."` unless that exact priority is explicitly present in `lead-sample.json`. Translate to low-certainty buyer context or omit the signal from copy.
-- **source-y signal narration** — reject `"saw you on..."`, `"saw you engaging with..."`, `"you commented on..."`, `"your LinkedIn activity..."`, `"you might not remember the thread..."`, `"found you through [source] and your role looked close..."`, or any line that makes the recipient feel watched unless the chosen archived motion is intentionally self-aware about the signal. For sender-owned LinkedIn post sources, a light first-person acknowledgment is allowed when row data proves a reaction/comment: `"appreciate you showing some love on my post about [topic]"` or `"thanks for showing support on my [topic] post"`. Do not name a comment unless comment text is present. Follow the acknowledgment with a soft relevance bridge before broad pain/product copy, e.g. `"figured this might be relevant if LinkedIn is becoming more of a GTM channel for [company]"`. For Shared Signal Discovery, prefer a high-level topic bridge when source evidence supports the topic: `"saw you around conversations about [topic], so hope this is relevant."` For non-evergreen third-party LinkedIn-post-sourced campaigns, a topic-level bridge is allowed when it explains why the note exists and stays apologetically uncertain: `"saw you in a few conversations about [topic], so may be off, but this seemed relevant."`, `"saw you in a few conversations around [topic], so hope this is relevant."`, or `"found you in a thread about [topic], so may be off, but this seemed relevant."` The topic must be a business theme, not raw source mechanics or activity-log detail. Shared Cold Fallback must not use this source language. Reserve `"raise your hand"` language for explicit lead-magnet comments, replies, or opt-ins. Translate the signal into natural buyer context or omit it.
+- **source-y signal narration** — reject `"saw you on..."`, `"saw you
+  engaging with..."`, `"you commented on..."`, `"your LinkedIn activity..."`,
+  `"you might not remember the thread..."`, `"found you through [source] and
+  your role looked close..."`, or any line that makes the recipient feel
+  watched unless the chosen archived motion is intentionally self-aware about
+  the signal. For sender-owned LinkedIn post sources, a light first-person
+  acknowledgment is allowed when row data proves a reaction/comment:
+  `"appreciate you showing some love on my post about [topic]"` or `"thanks for
+  showing support on my [topic] post"`. Do not name a comment unless comment
+  text is present. Follow the acknowledgment with a soft relevance bridge
+  before broad pain/product copy, e.g. `"figured this might be relevant if
+  LinkedIn is becoming more of a GTM channel for [company]"`. For Shared Signal
+  Discovery and non-evergreen third-party LinkedIn-post-sourced campaigns, a
+  topic-level bridge is allowed when it explains why the note exists and stays
+  apologetically uncertain: `"saw you might be interested in [topic], so hope
+  this is relevant."`, `"saw you in a few conversations about [topic], so may be
+  off, but this seemed relevant."`, `"saw you in a few conversations around
+  [topic], so hope this is relevant."`, or `"hope this is relevant if [topic] is
+  still on your plate."` Shared evergreen signal override: Dotwork-style `saw
+  you might be interested` can be used for Shared Signal Discovery when the
+  named Dotwork pattern applies, topic-level and evidence-backed; otherwise
+  prefer the conversation/topic bridge. `found you in a thread` remains blocked
+  because it exposes sourcing mechanics.
+  Shared Cold Fallback should use a concrete signal/problem
+  bridge or omit the source line. Reserve `"raise your hand"` language for
+  explicit lead-magnet comments, replies, or opt-ins. Translate the signal into
+  natural buyer context or omit it.
 - **sender-owned source rendered as third-party** — hard fail when the source post is sender-owned and the draft says `"found you in a thread"`, `"saw you in a thread"`, `"saw you in conversations"`, or `"saw you in a few conversations"` as if the post were third-party. For sender-owned sources, the copy must either use a light first-person acknowledgment plus a soft relevance bridge, or omit the source line. If multiple senders may send the campaign, or the final sender/source-owner match is ambiguous, do not assume a specific `my post` voice or name a specific sender; omit the sender-owned source line until row-level sender ownership is proven. Neutral low-certainty thread/source language is only for truly third-party sources.
 - **fake line-to-line continuity** — reject line stacks where the source acknowledgment, relevance bridge, product line, and CTA do not actually build on each other. Each line must make the next line feel earned. If two adjacent lines could be swapped, deleted, or joined with `"anyway"` without changing the meaning, the transition is fake. In sender-owned post campaigns, the chain should be: support on my post -> why this topic may matter for the company -> what the product/problem does about that same topic -> low-friction next step.
 - **assumptive title-fit opener** — reject `"Your [role] role at [company] looked close to this problem"` or `"looked close to this outbound campaign problem"`. This asserts fit from title/company. Keep the apologetic uncertainty instead: `"may be off, but if [workflow] is relevant to what you're working on..."`.
@@ -375,14 +401,11 @@ who scraped me."
 - `Hey {{first_name}}, I'm building [X] — [what it does in one plain
 sentence].`
 - `Hey {{first_name}}, hope this is relevant if [buyer context] is
-still on your plate.` Non-evergreen weak-signal only.
+still on your plate.` Shared Signal Discovery or non-evergreen weak-signal only.
 - `Hey {{first_name}}, may be off, but if [workflow] is relevant to what
 you're working on, this might be useful.`
 - `Hey {{first_name}}, saw you might be interested in [topic], so hope
-this is relevant.` Non-evergreen weak-signal only.
-- `Hey {{first_name}}, saw you around conversations about [topic], so
-hope this is relevant.` Preferred for Shared Signal Discovery when the
-topic is high-level and source-supported.
+this is relevant.` Dotwork Shared Signal Discovery or non-evergreen weak-signal only.
 - `Hey {{first_name}}, saw you in a few conversations around [topic],
 so hope this is relevant.` Shared Signal Discovery or non-evergreen weak-signal only.
 - `Hey {{first_name}}, found you in a thread about [topic], so may be

package/skills/create-evergreen-campaigns/SKILL.md

@@ -409,37 +409,30 @@ acceptable only when it writes a fresh durable per-lane receipt artifact as it
 works.
 When visible Codex app thread tools are unavailable but local Codex CLI is
 available, the accepted durable streaming-worker command shape is:
-`codex -a never -s danger-full-access -c model_reasoning_effort=xhigh exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> -`.
-The approval, sandbox, and reasoning-effort config flags must come before the
-`exec` subcommand for Codex CLI builds that expose `-a`/`-s`/`-c` only at top
-level. `--skip-git-repo-check` belongs after `exec` because current customer and VPS Codex CLI builds expose it
+`codex -a never -s danger-full-access exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> -`.
+The approval and sandbox flags must come before the `exec` subcommand for Codex
+CLI builds that expose `-a`/`-s` only at top level. `--skip-git-repo-check` belongs after `exec` because current customer and VPS Codex CLI builds expose it
 as an `exec` subcommand flag, not a top-level flag. Do not use
 `codex exec --ask-for-approval never ...` or `codex exec -a never ...`; those
 forms fail on current customer CLI installs. Pipe the lane packet prompt on
 stdin, require the worker to write `workerDispatch.receiptArtifactHint`, and
 pass exactly one lane packet per worker.
 When launching durable Codex CLI workers from an automation parent, pass an
-explicit supported worker model and explicit `xhigh` reasoning effort instead
-of relying on the Codex CLI defaults. Use the parent runtime model when known,
-for example:
-`codex -a never -s danger-full-access -c model_reasoning_effort=xhigh exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> -`.
+explicit supported worker model instead of relying on the Codex CLI default
+model. Use the parent runtime model when known, for example:
+`codex -a never -s danger-full-access exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> -`.
 In Codex CLI, the parent runtime model is visible in the run header as
 `model: <model-name>`. Copy that exact model string into child worker launches
 first. If the parent header says `model: gpt-5.5`, launch workers with
-`-m gpt-5.5` plus `-c model_reasoning_effort=xhigh`; do not invent or probe
-nearby aliases such as `gpt-5.3-codex`, `gpt-5.2`, `gpt-5-codex`,
-`codex-latest`, or `codex-mini-latest` before trying the exact parent model. A
-one-line probe must include the same `-c model_reasoning_effort=xhigh` override
-and count as supported only when it exits 0 and returns the requested output; a
-session header followed by a `not supported` error is rejected, not accepted.
-Do not rely on the Codex CLI default model or default reasoning effort; some
-customer and VPS installs default to unavailable model aliases or to `high`
-reasoning. `-m gpt-5.5` alone is not enough. If a child worker reports GPT 5.5
-with `high` reasoning, treat that as a launcher bug, relaunch with the explicit
-`xhigh` config before mutation, and do not ask the user to continue through the
-model-quality warning. If the parent cannot identify a supported worker model
-and launch it with `xhigh` reasoning, stop with
-`blocked: worker_model_unavailable` before mutation.
+`-m gpt-5.5`; do not invent or probe nearby aliases such as `gpt-5.3-codex`,
+`gpt-5.2`, `gpt-5-codex`, `codex-latest`, or `codex-mini-latest` before trying
+the exact parent model. A one-line probe must count as supported only when it
+exits 0 and returns the requested output; a session header followed by a
+`not supported` error is rejected, not accepted.
+Do not rely on the Codex CLI default model; some customer and VPS installs
+default to unavailable model aliases. If the parent cannot identify a supported
+worker model and a test `codex exec -m <worker-model>` launch is rejected, stop
+with `blocked: worker_model_unavailable` before mutation.
 When wrapping multiple local Codex CLI workers in a shell launcher, run the
 wrapper with `/bin/bash -lc` or another explicitly chosen portable shell. Do not
 rely on zsh-specific behavior. Do not use `status` as a shell variable because
@@ -451,7 +444,7 @@ Do not embed `<<'WORKER_PROMPT'` heredocs inside a single quoted or double
 quoted `/bin/bash -lc '...'` command string; nested quoting is brittle and can
 truncate the first worker before mutation. For multi-worker launchers, write one
 plain prompt file per lane under the current run directory, then start each
-worker with `codex -a never -s danger-full-access -c model_reasoning_effort=xhigh exec --skip-git-repo-check -m "$WORKER_MODEL" -C "$REPO" -o
+worker with `codex -a never -s danger-full-access exec --skip-git-repo-check -m "$WORKER_MODEL" -C "$REPO" -o
 "$worker_final_file" - < "$worker_prompt_file"`. Keep launcher shell variables
 double-quoted and keep the prompt heredoc only in a standalone script/prompt-file
 write step, not inside an already quoted shell argument. If the first launcher
@@ -557,7 +550,7 @@ launchers, write the same prompt body to `<worker-prompt-file>` and launch the
 worker with stdin redirected from that file:
 
 ```
-codex -a never -s danger-full-access -c model_reasoning_effort=xhigh exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> - <<'WORKER_PROMPT'
+codex -a never -s danger-full-access exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> - <<'WORKER_PROMPT'
 Use $sellable:create-campaign as the governing campaign workflow for this one lane worker.
 Use the evergreen plan/packet below only for lane scope, source metadata,
 postconditions, side-effect caps, and durable receipt proof.
@@ -574,8 +567,7 @@ Write the durable receipt to this exact path: <receiptArtifactPath>
 
 If your runtime exposes a goal tool, create or maintain this exact goal:
 complete action <actionId> only when <receiptArtifactPath> exists as valid JSON
-for this lane, or only for the accepted Post Engagers no-source blockers listed
-below when a blocked receipt has been written at that same path.
+for this lane or when a blocked receipt has been written at that same path.
 
 Create-campaign goal prompt (authoritative; paste verbatim from the matching
 createCampaignGoals[].goalPrompt):
@@ -603,11 +595,6 @@ Use mcp__sellable only. Do not use admin tools, direct DB, Prisma, SQL, web sear
 Do not launch, start, schedule, send, or use paid InMail.
 For filter proof, durable receipt status must be `filterDecisionReceipt.status:"applied"` only; never `completed`, `confirmed`, `done`, or aliases.
 For generated messages, `update_cell` is allowed only for the semantic Approved checkbox. Never use `update_cell` for generated message text/body/sample copy. Bad copy requires `revise_message_template_and_rerun` or brief/template revision plus Generate Message rerun. Any generated-message cell override is `blocked: generated_message_cell_override`.
-If `bootstrap_create_campaign.modelQuality.status === "warn"` because the child
-worker reports GPT 5.5 with `high` reasoning, that is a parent launcher
-configuration bug, not an operator approval path inside the worker. Stop before
-mutation, tell the parent to relaunch this lane with
-`-c model_reasoning_effort=xhigh`, and do not mark the worker goal complete.
 Complete only this lane. Do not end with narration only. Before your final
 response, run a local file-existence and JSON self-check for
 <receiptArtifactPath>. If the lane succeeded, write the canonical success
@@ -621,14 +608,12 @@ receipt, not a worker crash: use one of
 `post_engagers_no_sender_owned_posts`, or
 `post_engagers_source_author_mismatch`, include the source/readback evidence,
 and stop without scraping third-party authors or generating misleading warm
-copy. Any other blocked receipt is not a goal-complete success condition; it is
-a failed lane return that the parent must report or repair. The terminal success
-condition is filesystem proof that <receiptArtifactPath> exists with a success
-receipt, or an accepted Post Engagers no-source blocked/no-op receipt.
+copy. The terminal condition is filesystem proof that <receiptArtifactPath>
+exists.
 WORKER_PROMPT
 
 # Multi-worker launcher equivalent:
-codex -a never -s danger-full-access -c model_reasoning_effort=xhigh exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> - < <worker-prompt-file>
+codex -a never -s danger-full-access exec --skip-git-repo-check -m <worker-model> -C <repo> -o <worker-final-file> - < <worker-prompt-file>
 ```
 
 If any placeholder cannot be filled from the current plan, matching
@@ -878,11 +863,7 @@ by someone else. Use the explicit blocker codes
 `acceptedLaneBlockers`, so the parent can report the truthful no-op instead of
 retrying invalid source repair. This exception does not apply to shared lanes,
 bad generated copy, stale plans, target mismatch, missing receipt files, launch
-or schedule side effects, model-quality/preflight blockers, worker model
-availability blockers, or arbitrary worker failures. A blocked receipt outside
-the accepted Post Engagers source codes must not be treated as completion, must
-not be used to mark the worker goal complete, and must not let the parent report
-the evergreen command as complete.
+or schedule side effects, or arbitrary worker failures.
 
 All step proof objects must live under `createCampaignStepReceipt`. Do not put
 `createCampaignWorkflowReceipt`, `campaignBriefReceipt`,
@@ -1325,7 +1306,6 @@ Message, and verify current-revision sample messages before final completion.
      Short form: Do not pass the warm template into Shared Signal Discovery or Shared Cold Fallback.
      Short form: Shared lanes must reject generic product-summary samples.
      Short form: Shared lanes must reject three nearly identical samples.
-     Short form: Shared lanes must reject unsupported low-confidence relevance hedges.
      Short form: Shared lane message quality is owned by the packaged generate-messages prompt and validation; evergreen does not author source-thread bridge phrasing.
      Short form: Shared lanes must reject internal workflow vocabulary.
      Short form: revise or rerun the packaged create-campaign/generate-messages path.
@@ -1367,12 +1347,21 @@ Message, and verify current-revision sample messages before final completion.
      product summaries, or repeated structure are bad. If the first generated
      batch violates these rules, rerun, but record that as a repaired UAT path
      rather than the desired zero-shot path.
-     Shared Signal Discovery samples may use a supported topic bridge such as
-     "saw you around conversations about [high-level topic], so hope this is
-     relevant" when the topic is grounded in source evidence and not
-     activity-log phrasing. Translate source/tool terms into prospect-facing
-     business language; do not put internal workflow vocabulary such as
-     "Codex-style workflow", "MCP", "agent workflows",
+     Use the Dotwork gold-standard shape for Shared Signal Discovery samples
+     when the row supports a topic-level weak signal: "saw you might be
+     interested in [topic], so hope this is relevant", "saw you in a few
+     conversations about [topic], so hope this is relevant", or "saw you in a
+     few conversations around [topic], so hope this is relevant". These bridges
+     are allowed only when source evidence supports the topic and the line does
+     not claim a reaction, comment, sender-authored post, or other engagement
+     that is not proven. Do not use "found you in a thread" or other raw
+     source/discovery mechanics.
+     Shared Signal Discovery InMail subjects should follow the Dotwork plus-sign
+     A + B + C pattern: concrete topic + buyer outcome + useful artifact/picture,
+     for example "initiatives + business outcomes + one live picture".
+     Translate source/tool terms into prospect-facing business language; do not
+     put internal workflow
+     vocabulary such as "Codex-style workflow", "MCP", "agent workflows",
      "Claude Code", or "Claude-style agent workflows" in customer-facing
      message one. Also avoid generic product-process nouns such as "dashboard",
      "workflow", "source", "send-ready copy", "reviewed prospect lists",
@@ -1383,7 +1372,6 @@ Message, and verify current-revision sample messages before final completion.
      Short form: Shared Signal Discovery uses no self-intro such as "I'm building Sellable".
      Short form: Shared Signal Discovery uses no self-intro such as "We're building Sellable".
      Short form: Shared Signal Discovery uses no "I've put together".
-     Short form: Shared Signal Discovery may use "saw you around conversations about [topic], so hope this is relevant" when source-supported.
      Short form: topic-level bridge rules are owned by the packaged generate-messages prompt and validation, not by the evergreen wrapper.
      Short form: no internal workflow vocabulary such as "Codex-style workflow", "MCP", "agent workflows", "Claude Code", or "Claude-style agent workflows".
    - **Shared Cold Fallback** first-message style: cold, direct, and cohesive.
@@ -1400,15 +1388,14 @@ Message, and verify current-revision sample messages before final completion.
      Sellable/Codex/Claude-Code feature summary
      across prospects. This is a single-message cold path unless the approved
      create-campaign message workflow proves a better delivery format.
-     Do not open with "hope this is relevant" or a similar low-confidence
-     relevance hedge in Shared Cold Fallback. Do not use a repeated product-definition paragraph such
-     as "Sellable is a GTM system" across samples; every reviewed cold sample
-     should be specific enough to the role, company, or problem that the three
-     sample messages do not share the same body skeleton.
-     Do not use source-thread hedges such as "might be interested", "saw you
-     around conversations about...", or "saw you in a few conversations" in
-     Shared Cold Fallback. Start from a concrete role, company, or buyer-problem
-     observation instead.
+     Do not use source-thread hedges such as "hope this is relevant", "might be
+     interested", or "saw you in a few conversations" in Shared Cold Fallback.
+     Start from a concrete role, company, or buyer-problem observation instead.
+     Do not use a repeated
+     product-definition paragraph such as "Sellable is a GTM system" across
+     samples; every reviewed cold sample should be specific enough to the role,
+     company, or problem that the three sample messages do not share the same
+     body skeleton.
      Do not use mechanical Sellable process language such as "campaign idea into
      prospects, filters, messages, and sequence setup", "sequence-prospects",
      "dashboard overhead", "campaign map", "send-ready copy", or "reviewed
@@ -1425,8 +1412,7 @@ Message, and verify current-revision sample messages before final completion.
      Short form: Shared Cold Fallback uses no "I've put together".
      Short form: Shared Cold Fallback uses no opt-out shrug such as "If this isn't relevant, ignore me".
      Short form: Shared Cold Fallback uses no product-stack dump.
-     Short form: Shared Cold Fallback uses no low-confidence relevance hedge such as "hope this is relevant".
-     Short form: no "hope this is relevant", "might be interested", "saw you around conversations about", or "saw you in a few conversations" in Shared Cold Fallback; Shared Signal Discovery has the topic-level conversation-bridge exception above.
+     Short form: no "hope this is relevant", "might be interested", or "saw you in a few conversations" in Shared Cold Fallback; Shared Signal Discovery has the topic-level conversation-bridge exception above.
      Short form: no repeated product-definition paragraph such as "Sellable is a GTM system".
     - Customer-visible evergreen lanes must enable ICP filtering before Message
       Drafting. This applies to Post Engagers, Shared Signal Discovery, and
@@ -1527,18 +1513,13 @@ Message, and verify current-revision sample messages before final completion.
       or a product-stack dump that repeats the same
       Sellable/Codex/Claude-Code feature summary across prospects. For all
       shared lanes, reject generic product-summary samples, repeated
-      near-identical samples, unsupported low-confidence relevance hedges, internal
-      workflow vocabulary, any sample with non-empty `issues`, and any preserved
-      Post Engagers opener; revise or rerun the packaged
-      create-campaign/generate-messages path before reporting completion. Do not
-      approve or final-report samples with `verdict:"passed"` and non-empty
-      `issues`.
-      Shared Signal Discovery samples may use `"saw you around conversations
-      about [high-level topic], so hope this is relevant"` when the topic is
-      supported by source evidence and not activity-log phrasing. Shared Cold
-      Fallback samples must still reject source/conversation hedges such as
-      `"hope this is relevant"`, `"might be interested"`, or `"saw you in a few
-      conversations"`.
+      near-identical samples, intent-assumptive hedges such as "might be
+      interested", source-context mismatches such as "saw you in a few
+      conversations" in Shared Cold Fallback, internal workflow vocabulary, any
+      sample with non-empty `issues`, and any preserved Post Engagers opener;
+      revise or rerun the packaged create-campaign/generate-messages path before
+      reporting completion. Do not approve or final-report samples
+      with `verdict:"passed"` and non-empty `issues`.
    - The first review batch exists and at least 3 review rows have generated
      messages from the approved brief. If fewer than 3 usable rows exist, report
      the actual count and why.

package/skills/generate-messages/SKILL.md

@@ -246,12 +246,6 @@ says `generate-messages` loaded or because generated cells exist. The samples
 must read like final create-campaign/generate-messages output.
 
 - Start from a concrete row-supported role/company/problem observation.
-- Do not turn a company category into a certainty claim about the buyer's
-  channel strategy. Reject lines shaped like `Building [product/category] for
-  [buyers] means LinkedIn attention maps to pipeline`, `because you sell to
-  enterprise, LinkedIn is obviously a pipeline lever`, or other inferred
-  inevitability claims unless the row explicitly states that fact. A company
-  category can inform targeting; it is not buyer-facing proof.
 - Ask one clear buyer problem question.
 - Keep the product line minimal and buyer-readable; do not turn the note into a
   product-stack dump.
@@ -275,15 +269,15 @@ must be safer than the legacy weak-signal examples. These lanes are reused by
 multiple senders, so the first line cannot sound like a scraped-activity excuse
 or a founder self-intro.
 
-- Shared Signal Discovery may use a soft, high-level topic bridge when source
-  evidence supports the topic: `saw you around conversations about
-  {{signal_topic}}, so hope this is relevant`. Keep the topic broad and
-  business-relevant; do not recap the source, reaction, comment, post, author,
-  or scraper path.
-- Shared Cold Fallback must not use Signal Discovery source language such as
-  `saw you around conversations about...`, `saw you in a few conversations...`,
-  `might be interested`, or `hope this is relevant`. It is a true cold lane and
-  should start from a buyer problem or row-supported observation.
+- For Shared Signal Discovery, low-certainty topic bridges are allowed when
+  evidence-backed: `saw you might be interested in [topic], so hope this is
+  relevant`, `saw you in a few conversations about/around [topic], so hope this
+  is relevant`, and `hope this is relevant if [topic] is part of the plan`. The
+  line must not claim a reaction, comment, sender-authored post, or other
+  engagement that is not proven.
+- `found you in a thread` is blocked for Shared Signal Discovery because it
+  exposes source mechanics. Shared Cold Fallback should use concrete
+  role/company/problem context instead of source-thread phrasing.
 - do not write first-person singular founder language such as `I'm building`,
   `I've mapped`, `my company`, or `my team` unless exactly one sender can
   credibly say it and the lane is not shared.
@@ -294,18 +288,24 @@ or a founder self-intro.
   `source`, `sequence`, `send-ready copy`, or `reviewed prospect list` as
   buyer-facing explanation in shared evergreen samples unless the buyer's own
   company/product category uses that exact word and the line still sounds human.
-- Use a concrete signal/problem bridge only when it is less presumptive than the
-  topic bridge. Do not force a generic `LinkedIn signal into pipeline` line just
-  because the lane is shared.
-- If the row evidence is too weak to name a high-level topic, omit the source
+- Write a concrete signal/problem bridge instead of a rejected relevance hedge.
+  Good shapes are `the AI-assisted GTM conversations keep coming back to [buyer
+  problem]`, `teams trying to turn LinkedIn signal into pipeline usually hit
+  [problem]`, or `when [observable business moment] is live, [buyer workflow]
+  gets harder to run consistently`.
+- If the row evidence is too weak to write a concrete bridge, omit the source
   line and start from a buyer problem, event, role, or company observation.
 - Shared evergreen lanes must pass this gate before `confirmed`, before saving a
   reusable template, and before any route-proof approval.
 
-Short form: Shared Signal Discovery can use `saw you around conversations about {{signal_topic}}, so hope this is relevant`.
-Short form: Shared Cold Fallback cannot use Signal Discovery source language.
+Short form: shared evergreen copy must start from buyer context or an
+evidence-backed topic bridge, not assumed intent.
+Short form: Shared Signal Discovery may use `saw you might be interested in
+[topic], so hope this is relevant`, `saw you in a few conversations
+about/around [topic], so hope this is relevant`, or `hope this is relevant if
+[topic] is part of the plan` when it is topic-level and evidence-backed.
 Short form: shared evergreen copy cannot use `I'm building`.
-Short form: do not force generic LinkedIn-pipeline assumptions when a topic bridge is safer.
+Short form: write a concrete signal/problem bridge instead of a rejected relevance hedge.
 
 ### Rich Personalization Contract
 
@@ -325,15 +325,16 @@ conditional good/bad fill reference.
   `[PERSONALIZATION_LINE — Intent: write one short sentence that anchors the
 note to why this might matter now. DO: connect one observable role/company/
 activity signal to the buyer problem or offer with a permissioned bridge. For
-  Shared Signal Discovery, prefer `saw you around conversations about [high-level
-  topic], so hope this is relevant` when source evidence supports the topic.
-  For Shared Cold Fallback, do not use source/conversation language; start from a
-  buyer problem or row-supported observation. Legacy low-certainty language like
-  "so may be off, but this seemed relevant" can be considered for non-evergreen
-  weak-signal campaigns when the source is otherwise awkward. DON'T: recap their
-  resume, list companies, compliment their
-  background, name a product their company sells, use source-citation phrasing,
-  or use generic nouns like "your work".
+shared evergreen lanes, do not use source-thread disclaimers that expose
+mechanics. For Shared Signal Discovery, `saw you might be interested in [topic],
+so hope this is relevant`, `saw you in a few conversations about/around [topic],
+so hope this is relevant`, and topic-backed `hope this is relevant` bridges are
+allowed when topic-level and evidence-backed. Use a concrete buyer/problem
+bridge when that is stronger. Low-certainty language like
+"so may be off, but this seemed relevant" can be considered only for
+non-evergreen weak-signal campaigns when the source is otherwise awkward. DON'T: recap their resume, list companies, compliment their
+background, name a product their company sells, use source-citation phrasing, or
+use generic nouns like "your work".
 FALLBACK: if unsupported, omit the entire line.]`
 - The rendered personalization line should usually be 6-14 words. If it needs
   two clauses, multiple companies, or abstract tags like `GTM`, `international
@@ -485,19 +486,17 @@ artifact.**
    - `your {{role}} role at {{company}} looked close to this problem`
    - `your {{role}} role at {{company}} looked close to this outbound campaign problem`
    - `saw you on {{recent_signal}}`
-   - `saw you around {{recent_signal}}` when `recent_signal` is raw
-     post/source/activity text instead of a high-level topic
+   - `saw you around {{recent_signal}}`
    - `saw you engaging with {{recent_signal}}`
    - `saw you raise your hand for {{signal_topic}}, so figured this was worth sending`
    - `figured this was worth sending`
 
    Safer shapes:
 
-   - `hope this is relevant if {{signal_topic}} is part of the plan` for non-evergreen weak-signal campaigns
-   - `saw you might be interested in {{signal_topic}}, so hope this is relevant` for non-evergreen weak-signal campaigns
-   - `saw you around conversations about {{signal_topic}}, so hope this is relevant` for Shared Signal Discovery and other third-party weak-signal campaigns when the topic is high-level and supported by source evidence
-   - `saw you in a few conversations around {{signal_topic}}, so hope this is relevant` for Shared Signal Discovery and other third-party weak-signal campaigns when the topic is high-level and supported by source evidence
-   - `saw you in a few conversations about {{signal_topic}}, so may be off, but this seemed relevant` for non-evergreen weak-signal campaigns
+   - `hope this is relevant if {{signal_topic}} is part of the plan` for Shared Signal Discovery or non-evergreen weak-signal campaigns only
+   - `saw you might be interested in {{signal_topic}}, so hope this is relevant` for Shared Signal Discovery or non-evergreen weak-signal campaigns only
+   - `saw you in a few conversations around {{signal_topic}}, so hope this is relevant` for Shared Signal Discovery or non-evergreen weak-signal campaigns only
+   - `saw you in a few conversations about {{signal_topic}}, so may be off, but this seemed relevant` for non-evergreen weak-signal campaigns only
    - `appreciate you showing some love on my post about {{signal_topic}}` only when the sender/client authored the source post and the row proves a reaction or comment
    - `thanks for showing support on my {{signal_topic}} post` only when the sender/client authored the source post and the row proves a reaction or comment
    - `figured this might be relevant if {{channel_context}} is becoming more of a {{workflow_context}} for {{company_context}}` as the bridge after a sender-owned post acknowledgment
@@ -505,7 +504,7 @@ artifact.**
    - `if {{workflow_signal}} is not relevant to what you're working on, ignore me`
    - `figured this might matter if {{workflow_signal}} is live at {{company}}`
    - `thought of {{company}} because {{observable_signal}} points at {{problem_context}}`
-   - `found you in a thread about {{signal_topic}}, so may be off, but this seemed relevant` for non-evergreen weak-signal campaigns only
+   - `found you in a thread about {{signal_topic}}, so may be off, but this seemed relevant`
    - `saw you raise your hand for {{signal_topic}}, so figured this was (hopefully) worth sending` only when the source was an explicit lead-magnet comment, reply, or opt-in. The parenthetical uncertainty is load-bearing; do not flatten it to `so figured this was worth sending`.
    - `saw you raise your hand for {{signal_topic}} (creepy to reach out based on that, i know) - but this felt too on the nose to ignore` only when the source was an explicit lead-magnet comment, reply, or opt-in and the sender's voice can carry the cheekier aside
 
@@ -559,26 +558,27 @@ artifact.**
    For third-party LinkedIn-post-sourced campaigns (commenters, reactors, lead
    magnet replies, or scraped conversations), it is acceptable to reference the
    conversation when it explains why the note exists. Keep it topic-level, not
-   activity-log-level: `saw you around conversations about {{signal_topic}}, so
-   hope this is relevant` is preferred for Shared Signal Discovery when the
-   topic is high-level and supported by source evidence. `saw you in a few
-   conversations around {{signal_topic}}, so hope this is relevant` is also
-   acceptable when it reads more natural. `saw you commented on
-   {{post_context}}` and `your LinkedIn activity around...` are not acceptable.
-   Reserve `raise your hand` phrasing for explicit lead-magnet comments,
-   replies, or opt-ins. For Shared Cold Fallback, omit the source line and start
-   from a buyer problem or row-supported cold observation.
+   activity-log-level. `saw you commented on {{post_context}}` and `your
+   LinkedIn activity around...` are not acceptable. Reserve `raise your hand`
+   phrasing for explicit lead-magnet comments, replies, or opt-ins. For ordinary
+   shared evergreen thread discovery, Shared Signal Discovery may use
+   `saw you might be interested in {{signal_topic}}, so hope this is relevant`
+   or `saw you in a few conversations about/around {{signal_topic}}, so hope this
+   is relevant` when evidence-backed.
+   Shared Cold Fallback should translate the source topic into a concrete
+   buyer/problem bridge or omit the source line.
 
    Keep the apologetic nature, but aim it at uncertainty, not surveillance.
    Good for non-evergreen weak-signal campaigns: `saw you in a few
    conversations about outbound, so may be off, but this seemed relevant`, `may
    be off, but if outbound is relevant to what you're working on...`, or `hope
-   this is relevant if...`. For Shared Signal Discovery, prefer `saw you around
-   conversations about [high-level topic], so hope this is relevant` when the
-   source topic is supported. For Shared Cold Fallback, these source hedges are
-   blocked; use cold buyer-context copy instead. Bad:
-   `your role at {{company}} looked close...`, because it asserts fit from a
-   title and tells the buyer about themselves.
+   this is relevant if...`. For Shared Signal Discovery, topic-backed `hope this
+   is relevant`, `saw you might be interested in...`, and `saw you in a few
+   conversations about/around...` are allowed when topic-level and
+   evidence-backed. For
+   Shared Cold Fallback, use a concrete signal/problem bridge. Bad: `your role
+   at {{company}} looked close...`, because it asserts fit from a title and
+   tells the buyer about themselves.
 
    Block wordier source narration such as `you might not remember the thread,
    but I found you through a {{signal_topic}} discussion and your
@@ -1575,11 +1575,11 @@ Before `confirmed`, run the selected winner against these gates:
   the lead sample. If desire is inferred, use tentative framing like
   `hope this is relevant if...`, `saw you might be interested in...`,
   `saw you in a few conversations around...`, `figured this might matter if...`,
-  or `if you've ever wanted...`. For Shared Signal Discovery, the preferred
-  opener is a high-level topic bridge such as `saw you around conversations
-  about [topic], so hope this is relevant` when source evidence supports the
-  topic. For Shared Cold Fallback, do not use source/conversation language; use
-  a cold buyer-problem opener instead.
+  or `if you've ever wanted...`. Exception: for Shared Signal Discovery,
+  topic-backed `hope this is relevant`, `saw you might be interested in...`, and
+  `saw you in a few conversations around...` are allowed when topic-level and
+  evidence-backed. Other shared evergreen lanes should start from the
+  public business theme, row role/company context, or buyer problem.
 - **Buyer-first opener:** after the greeting, the first substantive line
   should usually enter the buyer's world: a real signal, current-state
   pain, buyer-owned priority, or useful artifact. Sender-origin can come
@@ -1875,18 +1875,18 @@ infrastructure signal`, or `rails can match` unless that exact buyer
 about`, or `clearly relevant` are BLOCKED. Source-y phrases such as
   `saw you on...`, bare `saw you around...`, and `saw you engaging with...`
   are also blocked unless the archived winner explicitly depends on that
-  self-aware signal style. For non-evergreen weak-signal campaigns, translate
-  the signal into low-certainty buyer context instead: `hope this is relevant if
-  [topic] is part of the plan`, `saw you might be interested in [topic], so hope
-  this is relevant`, `saw you in a few conversations around [topic], so hope
-  this is relevant`, `found you in a few conversations about [topic], so may be
-  off, but this seemed relevant`, `figured this might matter if [workflow] is
-  live at [company]`, or a similarly natural line. For Shared Signal Discovery,
-  keep the source bridge at high-level topic only:
-  `saw you around conversations about [topic], so hope this is relevant`. For
-  Shared Cold Fallback, skip the `hope this is relevant` / `might be interested`
-  / `saw you in a few conversations` versions and use cold buyer-context copy.
-  Longer source-recap forms such as `you
+  self-aware signal style. Translate the signal into low-certainty buyer
+  context instead: `hope this is relevant if [topic] is part of the plan`,
+  `saw you might be interested in [topic], so hope this is relevant`, `saw you
+  in a few conversations around [topic], so hope this is relevant`, `found you
+  in a few conversations about [topic], so may be off, but this seemed
+  relevant`, `figured this might matter if [workflow] is live at [company]`, or
+  a similarly natural line. For Shared Signal Discovery, topic-backed `hope this
+  is relevant`, `saw you might be interested ... hope this is relevant`, and
+  `saw you in a few conversations ... hope this is relevant` are allowed when
+  they stay topic-level and avoid false engagement claims. Other shared evergreen lanes should use
+  the concrete bridge version. Longer
+  source-recap forms such as `you
   might not remember the thread,
   but I found you through a [topic] discussion...` are BLOCKED; they narrate
   the scrape path instead of creating buyer relevance.
@@ -2032,11 +2032,10 @@ matters to the buyer, why this might be relevant, or the operator-reality that
 led to it. Acceptable openers:
 
 - `Hey {{first_name}}, [one plain sentence about what we built].`
-- `Hey {{first_name}}, hope this is relevant if [buyer context] is still on your plate.` Non-evergreen weak-signal only.
-- `Hey {{first_name}}, saw you might be interested in [topic], so hope this is relevant.` Non-evergreen weak-signal only.
-- `Hey {{first_name}}, saw you around conversations about [topic], so hope this is relevant.` Preferred for Shared Signal Discovery when source evidence supports a high-level topic.
-- `Hey {{first_name}}, saw you in a few conversations around [topic], so hope this is relevant.` Allowed for Shared Signal Discovery when it stays high-level and not activity-log-like.
-- `Hey {{first_name}}, found you in a thread about [topic], so may be off, but this seemed relevant.` Non-evergreen weak-signal only.
+- `Hey {{first_name}}, hope this is relevant if [buyer context] is still on your plate.` For Shared Signal Discovery or non-evergreen weak-signal campaigns only.
+- `Hey {{first_name}}, saw you might be interested in [topic], so hope this is relevant.` For Shared Signal Discovery or non-evergreen weak-signal campaigns only.
+- `Hey {{first_name}}, saw you in a few conversations around [topic], so hope this is relevant.` For Shared Signal Discovery or non-evergreen weak-signal campaigns only.
+- `Hey {{first_name}}, found you in a thread about [topic], so may be off, but this seemed relevant.`
 - `Hey {{first_name}}, saw you raise your hand for [topic], so figured this was (hopefully) worth sending.` Only use this for explicit lead-magnet comments, replies, or opt-ins.
 - `Hey {{first_name}}, saw you raise your hand for [topic] (creepy to reach out based on that, i know) - but this felt too on the nose to ignore.` Only use this for explicit lead-magnet comments, replies, or opt-ins.
 - Do not use the flattened line `so figured this was worth sending`; either keep
@@ -2296,12 +2295,11 @@ Rules:
 - if `lead-sample.json` carries **any** per-lead signal for this lead — a recent post, a recent hire, a visible tool in the headline, a topic engagement, a public initiative — the **message must reference it specifically**. Category-level copy is rejected.
 - default placement is **Block 3**, not Block 1. Keep Block 1 substance-first unless the archived motion clearly earns a signal-led opener without falling into source-citation / talk-at phrasing.
 - acceptable signal use by default:
-  - `"Hope this is relevant if [topic] is part of the plan."` Non-evergreen weak-signal only.
-  - `"Saw you might be interested in [topic], so hope this is relevant."` Non-evergreen weak-signal only.
-  - `"Saw you around conversations about [topic], so hope this is relevant."` Preferred for Shared Signal Discovery when source evidence supports a high-level topic.
-  - `"Saw you in a few conversations around [topic], so hope this is relevant."` Allowed for Shared Signal Discovery when it stays high-level and not activity-log-like.
-  - `"Saw you in a few conversations about [topic], so may be off, but this seemed relevant."` Non-evergreen weak-signal only.
-  - `"Found you in a thread about [topic], so may be off, but this seemed relevant."` Non-evergreen weak-signal only.
+  - `"Hope this is relevant if [topic] is part of the plan."`
+  - `"Saw you might be interested in [topic], so hope this is relevant."` Shared Signal Discovery or non-evergreen weak-signal only.
+  - `"Saw you in a few conversations around [topic], so hope this is relevant."` Shared Signal Discovery or non-evergreen weak-signal only.
+  - `"Saw you in a few conversations about [topic], so may be off, but this seemed relevant."` Shared Signal Discovery or non-evergreen weak-signal only.
+  - `"Found you in a thread about [topic], so may be off, but this seemed relevant."` Non-evergreen weak-signal only; blocked for shared evergreen lanes.
   - `"Thought of {{company}} because [observable signal] points at [problem]."`
   - `"Figured this might matter if [workflow] is live at {{company}}."`
   - `"Saw you raise your hand for [topic], so figured this was (hopefully) worth sending."` Only for explicit lead-magnet comments, replies, or opt-ins.
@@ -2316,26 +2314,32 @@ Rules:
   - `"Saw you engaging with [topic]."`
   - `"Saw you raise your hand for [topic], so figured this was worth sending."`
   - `"Figured this was worth sending."`
-- signal-led opener shape is a **special case**, not the default. Use it only when the archived motion truly depends on it and the line can stay natural. For Shared Signal Discovery, prefer `"saw you around conversations about [topic], so hope this is relevant"` when source evidence supports a high-level topic. For non-evergreen LinkedIn-post-sourced campaigns, `"saw you in a few conversations around [topic], so hope this is relevant"` or `"found you in a thread about [topic], so may be off, but this seemed relevant"` can be used when referencing the source makes the outreach feel less random. Avoid activity-log phrasing such as `"you commented on..."`, `"you reacted to..."`, or `"your LinkedIn activity..."` unless the approved motion is the deliberately self-aware raise-hand line for an explicit lead-magnet comment, reply, or opt-in.
-- For Shared Signal Discovery, prefer the high-level topic bridge above when it
-  is supported by source evidence. Use a concrete problem bridge only when it is
-  more specific and less presumptive than the topic bridge.
+- signal-led opener shape is a **special case**, not the default. Use it only when the archived motion truly depends on it and the line can stay natural. For Shared Signal Discovery and non-evergreen LinkedIn-post-sourced campaigns, source bridges such as `"saw you in a few conversations around [topic], so hope this is relevant"` or `"saw you in a few conversations about [topic], so may be off, but this seemed relevant"` can be used when referencing the source makes the outreach feel less random. Avoid activity-log phrasing such as `"you commented on..."`, `"you reacted to..."`, or `"your LinkedIn activity..."` unless the approved motion is the deliberately self-aware raise-hand line for an explicit lead-magnet comment, reply, or opt-in.
+- For Shared Evergreen Signal Discovery, topic-backed weak signal-led openers
+  such as `hope this is relevant if [topic] is part of the plan`, `saw you
+  might be interested in [topic], so hope this is relevant`, and `saw you in a
+  few conversations about/around [topic], so hope this is relevant` are allowed
+  when evidence-backed and not engagement-claiming. Translate the source into a
+  concrete problem bridge when that is stronger, e.g.
+  `"The AI-assisted GTM conversations keep coming back to one problem: turning
+  good LinkedIn signal into review-ready pipeline without another dashboard."`
+  Then connect the product in buyer language.
 - tokenized engagement-source opener shape is BLOCKED by default:
   `{{first_name}}, saw you {{engagement_context}} on {{post_context}}` reads
   like surveillance unless the filled line is explicitly self-aware and
-  low-certainty. For Shared Signal Discovery, prefer `saw you around
-  conversations about [topic], so hope this is relevant`. For non-evergreen
-  weak-signal campaigns, use `hope this is relevant if [topic] is part of the
-  plan`, `saw you might be interested in [topic], so hope this is relevant`,
-  `saw you in a few conversations around [topic], so hope this is relevant`,
-  `saw you in a few conversations about [topic], so may be off, but this seemed
-  relevant`, or `found you in a thread about [topic], so may be off, but this
-  seemed relevant`. For Shared Cold Fallback, those weak signal bridges are
-  blocked; use cold buyer-context copy instead. Use the self-aware raise-hand
-  shapes only when the source was an explicit lead-magnet comment, reply, or
-  opt-in: `saw you raise your hand for [topic], so figured this was (hopefully)
-  worth sending` or `saw you raise your hand for [topic] (creepy to reach out
-  based on that, i know) - but
+  low-certainty. Prefer a concise bridge such as `hope this is relevant if
+[topic] is part of the plan`, `saw you might be interested in [topic], so hope
+this is relevant`, `saw you in a few conversations around [topic], so hope this
+is relevant`, `saw you in a few conversations about [topic], so may be off,
+but this seemed relevant`, or `found you in a thread about [topic], so may be
+off, but this seemed relevant`. For Shared Signal Discovery, topic-backed `hope
+this is relevant`, `saw you might be interested ... hope this is relevant`, and
+`saw you in a few conversations ... hope this is relevant` are allowed when
+topic-level and evidence-backed. Other shared evergreen lanes should use a concrete signal/problem bridge. Use
+the self-aware raise-hand shapes only when
+the source was an explicit lead-magnet comment, reply, or opt-in: `saw you raise
+your hand for [topic], so figured this was (hopefully) worth sending` or `saw you
+raise your hand for [topic] (creepy to reach out based on that, i know) - but
 this felt too on the nose to ignore`. Do not use the longer `you might not remember the thread...` shape;
 if the concise source bridge still feels awkward, omit the engagement reference.
 - do not describe the sender/client in third person inside the outbound message.

package/skills/refill-sends-workflow/SKILL.md

@@ -447,26 +447,39 @@ Short form: all eligible enrolled senders are covered by sender-scoped `--yolo`;
 Short form: direct scheduler writes are not covered by `--yolo`.
 Short form: stop and re-plan instead of auto-accepting the changed action.
 
+Before choosing a refill prep primitive, inspect the table frontier from
+`get_campaign_refill_state.preparationFrontier`. If
+`hasLaterPreparedIsland:true` or `earliestUnpreparedRow` exists before later
+successful enrichment, treat that as a top-down gap: use
+`select_campaign_cells`/`queue_campaign_cells` with
+`rowSelector:{ type:"needsEnrichment" }` and `columnRole:"enrich"` in bounded
+row-position order, or use `start_campaign_message_preparation` with the
+adaptive defaults for the full enrich -> rubric -> Generate Message loop. Do not
+use the UI Jump anchor as the automation cursor; Jump is a navigation affordance
+and can be fooled by a later enriched island.
+
 After same-campaign source copy, inspect the actual campaign-table position of
-the copied rows before starting prep. Newly copied rows often land after the first 100 table rows. Use `get_campaign_table_schema` to read the `reviewBatch`,
-then use `select_campaign_cells` diagnostics against
-`rowSelector:{ type:"reviewBatch" }` for `enrich`, `generateMessage`, and
-`approved` roles when needed.
-
-For bounded split refills, operate on the review batch first. Queue
-`queue_campaign_cells` for `columnRole:"enrich"` on
-`rowSelector:{ type:"reviewBatch" }`, wait for campaign processing, then queue
-`columnRole:"generateMessage"` for the same review batch or
-`needsGeneratedMessage` rows. If the bounded packet explicitly uses
-`approvalMode:"approve"` because the user asked to fill/schedule sends or
-provided `--yolo` for a two-day send fill, approve only the bounded review-batch
-cohort after current generated messages exist. Do not broad approve existing
-table rows.
-
-If review-batch enrich cells are pending or Generate Message cells are
-dependency-blocked, omit `maxRowsToCheck` so the adaptive prep job can cover the
-bounded table scan, or set a bounded cap high enough to include appended rows
-when row count evidence requires it. Do not run a fixed `maxRowsToCheck:100` prep pass after appending rows to a larger existing campaign table; it can spend
+the copied rows before starting prep. Newly copied rows often land after the
+first 100 table rows. Use `get_campaign_table_schema` to read the `reviewBatch`
+and `select_campaign_cells` diagnostics against
+`rowSelector:{ type:"reviewBatch" }` only to understand the copied row set. Do
+not operate on `reviewBatch` before earlier `needsEnrichment` rows unless the
+approved packet explicitly says to prioritize the just-copied bounded split and
+the earlier rows are exhausted, dependency-blocked, or intentionally excluded.
+
+For bounded split refills, prefer top-down `needsEnrichment` or adaptive
+`start_campaign_message_preparation` first. If the bounded packet explicitly
+uses `approvalMode:"approve"` because the user asked to fill/schedule sends or
+provided `--yolo` for a two-day send fill, inspect approval candidates with
+`rowSelector:{ type:"needsApproval" }` after current generated messages exist,
+then approve only the bounded cohort covered by the packet. Do not broad approve
+existing table rows.
+
+If enrich cells are pending or Generate Message cells are dependency-blocked,
+omit `maxRowsToCheck` so the adaptive prep job can cover the bounded table scan,
+or set a bounded cap high enough to include the necessary row-position frontier
+when row count evidence requires it. Do not run a fixed `maxRowsToCheck:100`
+prep pass after appending rows to a larger existing campaign table; it can spend
 the whole budget on older rows and miss the refill batch.
 
 Bound `targetPreparedMessages` to the actual ready-to-schedule gap for the