@sellable/mcp 0.1.212 → 0.1.214

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sellable/mcp",
-  "version": "0.1.212",
+  "version": "0.1.214",
   "type": "module",
   "description": "Sellable MCP server for Claude Code and Codex campaign workflows",
   "main": "dist/index.js",

package/skills/create-campaign/SKILL.md

@@ -66,8 +66,8 @@ allowed-tools:
 
 Use this as the customer-facing public wrapper for Sellable campaign creation.
 It bootstraps auth and host capabilities, then loads the internal
-`create-campaign-v2` workflow prompt and `core/flow.v2.json` through MCP. Keep
-this wrapper thin; v2 is the operational source of truth.
+the internal campaign workflow prompt and `core/flow.v2.json` through MCP. Keep
+this wrapper thin; the packaged workflow is the operational source of truth.
 
 CampaignOffer state and the watch link are the customer-facing source of truth.
 Disk artifacts are optional debug/UAT diagnostics; normal customer runs should
@@ -722,7 +722,7 @@ updates.
      No other lines. No "all set", no "signed in", no other acknowledgement.
 
      After the user pastes the LinkedIn profile URL or handle, proceed with the
-     identity-first campaign setup in the v2 subskill prompt. Normalize handles
+     identity-first campaign setup in the internal workflow prompt. Normalize handles
      to a full profile URL, resolve it with `fetch_linkedin_profile`, and mark
      the client/company research gate with `complete_sender_research` when that
      protocol is required.
@@ -760,11 +760,10 @@ updates.
    Drafting whenever the host exposes it
    and the current host policy allows agent launch. The worker must load the
    full `mcp__sellable__get_subskill_prompt({ subskillName: "generate-messages" })`
-   prompt, every referenced message asset through
-   `mcp__sellable__get_subskill_asset`, and before returning
+   prompt, every required message asset named by `generate-messages` Mode 0
+   through `mcp__sellable__get_subskill_asset`, and before returning
    `mcp__sellable__get_subskill_prompt({ subskillName: "create-campaign-v2-validation" })`
-   plus any validation assets it references through
-   `mcp__sellable__get_subskill_asset` as the internal validation gate.
+   as the final internal validation gate.
    In Codex, YOLO/autonomous mode counts as campaign-scoped permission to use
    this single Message Drafting background agent. If the user has not enabled
    YOLO and has not explicitly asked for background agents, subagents, parallel
@@ -776,7 +775,7 @@ updates.
    silently fall back to parent-thread message drafting; parent fallback is
    allowed only when the user explicitly says to continue without a background
    agent.
-   Do not use any alternate or examples-only message prompt. Message review and
+   Do not use any alternate, examples-only, or local-artifact message prompt. Message review and
    message QA require Message Drafting output:
    do not draft from a checklist, local markdown artifact, or parent-thread
    intuition. Use campaign state, campaign brief content, selected source state, and
@@ -785,10 +784,10 @@ updates.
    synthesize local validation artifacts from general knowledge. The handoff to
    Message Drafting should pass lean basis only, not hashes, counts, or a long
    row-id list; the branch loads current brief/context, the full
-   `generate-messages` prompt, all referenced message assets,
-   `create-campaign-v2-validation`, and any validation assets it references. Do
-   not render fallback sample, concerns, or a QA receipt on the normal happy
-   path.
+   `generate-messages` prompt, all required message assets, and
+   `create-campaign-v2-validation`. The message recommendation handoff is
+   labeled Markdown, not raw JSON. Do not render fallback sample, concerns, or a
+   QA receipt on the normal happy path.
 6. Create the campaign shell early with the v1 brief so the user can open the
    watch link and see useful setup state immediately. Materialize the approved
    source list, copy confirmed rows into the campaign, and internally process the

package/skills/create-campaign/context/learnings.md

@@ -4,7 +4,7 @@ This log captures what has worked in real campaigns.
 
 Entries should be short, specific, and based on observed outcomes.
 
-**Cross-skill learnings** that would help other skills (engage, create-post) should ALSO be written to `.sellable/insights/cross-skill.md` so the entire system learns.
+**Cross-skill learnings** that would help other skills (engage, create-post) should ALSO be written to `~/.sellable/insights/cross-skill.md` so the entire system learns.
 
 ## Log
 

package/skills/create-campaign/references/brief-template.md

@@ -41,8 +41,8 @@ Example: "In a 15-minute call, we will audit your top 10 accounts and show where
 - **Bridge to:** [How it connects to their pain]
 - **Avoid:** [Generic phrases to skip - "best-in-class", "industry-leading", etc.]
 - **Voice:** [Reference core memory / composed engage memory first. Use legacy writing memory only as fallback.]
-- **Outbound rules:** [Reference `.sellable/configs/writing/outbound.md` if loaded -- observation -> relevance -> ask]
-- **Proof boundaries:** [Reference `.sellable/configs/core/proof-ledger.md` and `.sellable/configs/core/wins-ledger.md` if loaded. Do not upgrade or invent claims.]
+- **Outbound rules:** [Reference `~/.sellable/configs/writing/outbound.md` if loaded -- observation -> relevance -> ask]
+- **Proof boundaries:** [Reference `~/.sellable/configs/core/proof-ledger.md` and `~/.sellable/configs/core/wins-ledger.md` if loaded. Do not upgrade or invent claims.]
 
 ## Do NOT Target
 

package/skills/create-campaign-brief/references/brief-template.md

@@ -15,7 +15,7 @@ assistant message directly from the `result.result` stream-json payload —
 it does not read a file off disk.
 
 If you find yourself emitting a TodoWrite step like "Write brief.md v1
-to disk" or "Save brief to .sellable/..." — DELETE that step and return
+to disk" or "Save brief to ~/.sellable/..." — DELETE that step and return
 the brief inline instead. The tools `Write`, `Edit`, and `Bash` are
 intentionally not in the Phase 83 harness allowlist; attempting them
 will cause a ToolSearch loop that deadlocks the turn.
@@ -37,7 +37,7 @@ offer is, why it should work, and what must be true before we launch. Do not
 prescribe message structure, opener order, or final copy here.]
 
 Draft path:
-`.sellable/create-campaign-brief/drafts/{workspace-slug}/{campaign-slug}/brief.md`
+`~/.sellable/create-campaign-brief/drafts/{workspace-slug}/{campaign-slug}/brief.md`
 
 ## Product
 

package/skills/create-campaign-brief/references/draft-lifecycle.md

@@ -2,7 +2,7 @@
 
 Phase 83 writes one Markdown file:
 
-`.sellable/create-campaign-brief/drafts/{workspace-slug}/{campaign-slug}/brief.md`
+`~/.sellable/create-campaign-brief/drafts/{workspace-slug}/{campaign-slug}/brief.md`
 
 ## Rules
 

package/skills/create-campaign-brief/references/examples/briefs/gelee.md

@@ -58,7 +58,7 @@ _Personalization: References their previous exit or current venture by name + th
 
 Hi {{first_name}}, let me know if i'm off, but {{seems_like_you}}.
 
-saw you were active in some {{signal_topic}} conversations recently and figured this might be relevant.
+saw you in a few conversations around {{signal_topic}}, so hope this is relevant.
 
 we work with about 30 exited founders right now (yc, techstars, sequoia backed) handling their entire linkedin with ai + human review.
 
@@ -131,7 +131,7 @@ is this worth a look or should i send a short overview?
 
 Hi sarah, let me know if i'm off, but seems like you started building something new in the pharma space after the Benchling exit.
 
-saw you were active in some founder-led growth conversations recently and figured this might be relevant.
+saw you in a few conversations around founder-led growth, so hope this is relevant.
 
 we work with about 30 exited founders right now (yc, techstars, sequoia backed) handling their entire linkedin with ai + human review.
 

package/skills/create-campaign-brief/references/examples/briefs/superpower.md

@@ -35,70 +35,82 @@ Signal-based outreach using peer-to-peer tone. Positioning Superpower as early d
 
 **VARIANT 1 — Subject: jeff/{{first name}} intro**
 
-hey {{first name}}, saw you were active around some {{topic}} stuff recently and figured this might actually be relevant
+hey {{first name}},
 
-any chance you're exploring early detection or preventive health programs at {{company}}?
+saw you in a few conversations around {{topic}}, so hope this is relevant.
 
-we screen for 1,000+ conditions from a single blood draw and surface risks before they ever become claims.
+if preventive health is anywhere near the benefits plan at {{company}}, Superpower helps the team spot risk earlier.
 
-basically your team gets to see what's coming instead of just reacting to it
+it screens for 1,000+ conditions from a single blood draw and surfaces risks before they become claims.
 
-can I send over some more info?
+basically your team gets to see what's coming instead of only reacting to it.
+
+should i send over the short version?
 
 **VARIANT 2 — Subject: jeff/{{first name}} intro**
 
-hey {{first name}}. saw you were active in some {{topic}} conversations recently and wanted to run something by you
+hey {{first name}},
+
+saw you in a few conversations around {{topic}}, so hope this is relevant.
 
 most annual physicals test 15 biomarkers max.
 
-there's over 100 that actually matter and the gap between those two numbers is where the expensive claims tend to sit
+Superpower looks at 100+ biomarkers and screens for 1,000+ conditions from one draw.
 
-teams like Lyft and Notion are already exploring this. one draw, 1,000+ conditions, $199 a person.
+teams like Lyft and Notion are already exploring this at $199 a person.
 
-if your claims costs have been going up and you're not sure where it's coming from... that gap may be a good place to start looking
+if preventive health is nowhere near your lane, ignore me.
 
-can I send over some more info?
+should i send over the short version?
 
 ## Examples
 
 **Example 1 (Variant 1 — self-insured health plans):**
 Subject: susana/prab intro
 
-hey susana, saw you were active around self-insured health plan content recently and figured this might actually be relevant
+hey susana,
+
+saw you in a few conversations around self-insured health plans, so hope this is relevant.
 
-any chance you're exploring preventive health programs at Quantum Health?
+if preventive health is anywhere near the benefits plan at Quantum Health, Superpower helps the team spot risk earlier.
 
-we screen for 1,000+ conditions from a single blood draw and surface risks before they ever become claims.
+it screens for 1,000+ conditions from a single blood draw and surfaces risks before they become claims.
 
-basically your team gets to see what's coming instead of just reacting to it
+basically your team gets to see what's coming instead of only reacting to it.
 
-can I send over some more info?
+should i send over the short version?
 
 **Example 2 (Variant 2 — employee wellness programs):**
 Subject: lisa/prab intro
 
-hey lisa. saw you were active in some employee wellness programs conversations recently and wanted to run something by you
+hey lisa,
+
+saw you in a few conversations around employee wellness programs, so hope this is relevant.
 
 most annual physicals test 15 biomarkers max.
 
-there's over 100 that actually matter and the gap is driving the surprise expensive claims every year
+Superpower looks at 100+ biomarkers and screens for 1,000+ conditions from one draw.
+
+teams like Lyft and Notion are already exploring this at $199 a person.
 
-teams like Lyft and Notion are already exploring this. one test, 1,000+ conditions, $199 a person
+if preventive health is nowhere near your lane, ignore me.
 
-can I send over some more info?
+should i send over the short version?
 
 **Example 3 (Variant 1 — employee wellness programs):**
 Subject: princess/prab intro
 
-hey princess, saw you were active around employee wellness programs content recently and figured this might actually be relevant
+hey princess,
+
+saw you in a few conversations around employee wellness programs, so hope this is relevant.
 
-any chance you're exploring preventive health programs at Marriott?
+if preventive health is anywhere near the benefits plan at Marriott, Superpower helps the team spot risk earlier.
 
-we screen for 1,000+ conditions from a single blood draw and surface risks before they ever become claims.
+it screens for 1,000+ conditions from a single blood draw and surfaces risks before they become claims.
 
-basically your team gets to see what's coming instead of just reacting to it
+basically your team gets to see what's coming instead of only reacting to it.
 
-can I send over some more info?
+should i send over the short version?
 
 ## Social Proof
 
@@ -112,7 +124,8 @@ Teams like Lyft and Notion are exploring Superpower's early detection platform.
 - **ICP filtering:** All prospects filtered against ICP criteria before entering send queue
 - **Positioning:** Early detection tool for claims prevention, not a wellness benefit
 - **Tone:** All lowercase, casual, peer-to-peer - building relationship before asking
-- **Two approved variants:** Testing different hooks (exploratory vs. problem-aware)
+- **Two approved variants:** Testing a low-certainty signal bridge vs. proof-led
+  benefits mechanism
 
 ### Rubric Markdown
 
@@ -212,8 +225,8 @@ Teams like Lyft and Notion are already using early detection screening. One bloo
 
 **Subject: Prab/{{first_name}} intro**
 
-Hey {{first_name}}, not sure if you've looked into this but felt it may be relevant to your work.
+Hey {{first_name}},
 
-have you ever looked into early detection or preventive health programs for {{company}}?
+hope this is relevant if preventive health is anywhere near the benefits plan at {{company}}.
 
-we screen for 1,000+ conditions from a single blood draw and surface risks before they ever become claims.
+Superpower screens for 1,000+ conditions from a single blood draw and surfaces risks before they become claims.

package/skills/create-campaign-brief/references/phase75-active-runtime-message-pack.md

@@ -1,4 +1,4 @@
-# Phase 75 Active Runtime Message Pack
+# Active Runtime Message Pack
 
 This is the narrow message inspiration pack for `create-campaign-brief`.
 
@@ -14,7 +14,8 @@ Included as primary gold examples:
 - `Galley`
 - `Clover`
 - `Persona`
-- `Superpower`
+- `Superpower` — domain-language reference only; use the cleaned low-certainty
+  opener, not retired engagement-source phrasing
 
 Not included in runtime inspiration:
 
@@ -33,6 +34,9 @@ Not included in runtime inspiration:
 
 - Feed only endorsed examples into generation.
 - Keep bad lines only if they are explicitly labeled `BAD` and paired with a `BETTER` version.
+- Treat sellable.dev, Hey Digital, Persona, Galley, and Clover as the primary
+  line-level quality anchors. Superpower is useful for benefits/healthcare
+  mechanism language, but it must not outrank those examples on opener style.
 - Default to one approved message in the generated brief unless the ICP is genuinely broad.
 - A/B CTA is optional. Use it only when the second option is truly strong.
 - If there is a great low-friction proof asset or case study, that is often the best Option B.
@@ -134,22 +138,26 @@ b) I send you our B2B Ads Arsenal — ad templates, playbooks, and case studies
 
 Why included:
 
-- strongest concrete operational offer
-- easy to picture the output
-- why it works: the pain is tactile, the mechanism is easy to visualize, and the
-  CTA lets the buyer test the product with their own real input instead of
-  committing to a generic call
+- strongest signal-led honest-operator foodservice motion
+- concrete buyer pain before product category
+- why it works: the source bridge is low-certainty, the mechanism connects real
+  operating objects, and the CTA names a useful conversation instead of asking
+  for a generic call
 
 ```md
-hey {{first_name}},
+Hey Mohamed,
+
+Found you in a thread about food cost margins, so may be off, but this seemed relevant.
 
-this might not apply but if you're still rebuilding menu plans by hand every time a rotation changes, there's a way to skip that whole step.
+Bulk costing gets messy when recipes, portions, and vendor prices live apart.
 
-we built a tool that takes a menu from whatever format it's in and just... builds the whole plan. recipes, purchasing, stations, all connected. no more starting from scratch every cycle.
+Galley is a Culinary Resource Planning platform for foodservice teams.
 
-works best for {{company_team_type}} running rotations across a bunch of locations where every menu change turns into hours of work nobody has time for.
+It keeps recipes, ingredient costs, and menu margins connected - where purchasing and production can work from the same numbers.
 
-easiest way to see if it's worth a look is to try it with one of your actual menus. happy to set that up or I can send a 2-min video instead.
+Open to a quick call on recipe costing and margin visibility?
+
+P.S. 2,400+ kitchens already use Galley - with a 4.9 average rating :)
 ```
 
 ### Clover
@@ -213,36 +221,19 @@ Why included:
 
 - simple language in a hard domain
 - useful plain-language healthcare constraint example
-- why it works: it asks about the buyer's existing benefits priority, explains
-  the mechanism in plain language, and keeps the ask tiny instead of forcing a
-  call
+- why it works: it uses a low-certainty signal bridge, explains the mechanism
+  in plain language, and keeps the ask tiny instead of forcing a call
 
 ```md
 Hey {{first_name}},
 
-not sure if you've looked into this but felt it may be relevant to your work.
-
-have you ever looked into early detection or preventive health programs for {{company}}?
-
-we screen for 1,000+ conditions from a single blood draw and surface risks before they ever become claims.
-
-basically your team gets to see what's coming instead of just reacting to it
-
-can I send over some more info?
-```
-
-Locked production variant from the same Superpower campaign:
-
-```md
-subject: {{company}} + 1,000 biomarkers + 24/7 care team
-
-hey {{first_name}},
+saw you in a few conversations around {{topic}}, so hope this is relevant.
 
-curious how you're thinking about preventative health + benefits innovation given your work at {{company}}.
+if preventive health is anywhere near the benefits plan at {{company}}, Superpower helps the team spot risk earlier.
 
-we provide companies direct access to 100+ biomarker lab testing and screen 1,000+ conditions for every employee.
+it screens for 1,000+ conditions from a single blood draw and surfaces risks before they become claims.
 
-then we drive behaviour change with a 24/7 care team.
+basically your team gets to see what's coming instead of only reacting to it.
 
-open to a 15 min call to learn more, or should i send over a short overview first?
+should i send over the short version?
 ```

package/skills/create-campaign-v2/SKILL.md

@@ -4,9 +4,9 @@ description: compact JSON-gated shell-first Sellable campaign flow through sourc
 visibility: internal
 ---
 
-# Create Campaign v2
+# Create Campaign Workflow
 
-fast entry point. Run the active state machine in `core/flow.v2.json`;
+Fast entry point. Run the active state machine in `core/flow.v2.json`;
 load deeper references only at the stage that needs them.
 Load the compact flow for exact gates:
 
@@ -27,6 +27,10 @@ handoff read campaign state first: `campaignId`, `watchUrl`, `campaignBrief`,
 `leadScoringRubrics`, `approvedMessageTemplate`, `senderIds`,
 `sequenceTemplate`, and running state. Local draft files are debug-only.
 
+There is no normal approval-packet, commit-gate, or atomic-mint path in the
+active flow. Legacy packet/mint artifacts stay in
+`create-campaign-v2-validation` and rehearsal material only.
+
 ## Normal Flow
 
 1. Bootstrap and tell the user the active Sellable workspace.
@@ -305,7 +309,7 @@ background.
 
 Codex: YOLO counts as permission for this one post-import worker. If no
 permission, ask once after filter choice. Use generic `gpt-5.5` / `xhigh`
-Message Drafting if needed. Never draft in parent silently.
+Message Drafting if needed.
 
 If the user critiques or edits the template before `approve-message`, route the
 feedback back to Message Drafting with current `messageDraftRecommendation`,
@@ -318,18 +322,21 @@ and validation before approval also belong to Message Drafting, but validation
 is an internal gate. Do not render `renderedFallbackSample`, `concerns`, or a
 full `qaReceipt` in the normal happy path.
 
-Load:
+Run `post-find-leads-message-scout` as Message Draft Builder when exposed. It
+must load:
 
 ```text
 get_subskill_prompt({ subskillName: "generate-messages" })
-get_subskill_asset(...) for every asset in generate-messages Reference Asset Loading
+get_subskill_asset(...) for every required message asset named by generate-messages Mode 0
 get_subskill_prompt({ subskillName: "create-campaign-v2-validation" })
 ```
 
-Do not use any alternate or examples-only message prompt. Do not render message
-review until `messageDraftRecommendation` proves current campaign/table basis
-and shows whether it came from Message Drafting or an explicitly approved inline
-fallback.
+After candidate generation/revision and before returning `ready`, Message Draft
+Builder must load `create-campaign-v2-validation` through MCP as the final gate.
+No alternate, examples-only, or local-artifact prompt. If the agent cannot
+launch, the parent fallback runs the same gate from live state. Render review
+only after `messageDraftRecommendation` proves current campaign/table basis.
+Handoff: labeled Markdown, not raw JSON.
 
 ## Hard Gates
 

package/skills/create-campaign-v2/core/auto-execute.README.md

@@ -2,7 +2,7 @@
 
 This file narrates `auto-execute.yaml`. The YAML is the source of truth; this
 README explains intent, tuning guidance, and how each knob affects the
-Plan 85-02 autonomous tail.
+autonomous tail.
 
 ## When this file is loaded
 
@@ -24,7 +24,7 @@ into the campaign).
   confirmed source rows into the campaign. The tail MUST NOT exceed this cap
   before the message template is approved and at least one generated passing
   message is reviewed.
-- **`provider`** — Inherited from the Phase 84 commit decision (saved in
+- **`provider`** — Inherited from the approved source decision (saved in
   the committed brief / campaign metadata). Present here so the tail can
   narrate which provider it's running against; it does NOT override the
   committed provider.
@@ -54,8 +54,8 @@ into the campaign).
   proceeds.
   Default `strict` because live campaign state shouldn't ship messages
   with unresolved templating.
-- **`critique.enabled`** — Plan 85-03 feature flag for the message
-  critique pass. Plan 85-02 reads the flag but never flips it; it
+- **`critique.enabled`** — Feature flag for the optional message
+  critique pass. The plain campaign tail reads the flag but never flips it; it
   stays false by default. Only flip to `true` after dry-running the
   critique protocol against a real sample and confirming the token
   contract still holds on the rewritten output. See
@@ -77,7 +77,7 @@ into the campaign).
   voice). Each critic returns structured JSON the synthesis step can
   merge. Do NOT add free-form critics; stay on the three groups.
 - **`critique.synthesis.enforceFinalizerPass`** — When `true`, the
-  synthesizer re-runs the Phase 84 finalizer pass on its rewrite.
+  synthesizer re-runs the offline-validation finalizer pass on its rewrite.
   This is the last guardrail against critic-introduced token drift.
   Default stays `true`.
 - **`critique.opus.enabled`** — Flag gating Opus refinement
@@ -154,20 +154,20 @@ logs and adjust:
   conditions that short-circuit the tail.
 - `references/final-handoff-contract.md` — the dual greenlight channel
   that takes Step 16 into a running campaign.
-- `references/parallel-critique-protocol.md` — Plan 85-03 Step 15
+- `references/parallel-critique-protocol.md` — optional Step 15
   critique protocol: bounded sample, parallel critics, synthesis,
   budget caps, and token-contract enforcement.
-- `references/thomas-variant-selection.md` — Plan 85-03 rules for
+- `references/thomas-variant-selection.md` — rules for
   which rows earn an Opus refinement rewrite.
-- `references/thomas-revision-filters.md` — the positive-target and
+- `references/sellable-cleanup-rules.md` — the positive-target and
   automatic-revision filters that gate any critique rewrite.
 
 ## Flipping `messaging.critique.enabled` to true (dry-run protocol)
 
 The critique flag stays `false` by default because an enabled critique
 pass invents new tokens + costs ~$2 per sample pass — both of which
-can silently degrade a live campaign if the protocol isn't solid. Plan
-85-02 ships the protocol; flipping the default requires dry-run
+can silently degrade a live campaign if the protocol isn't solid. Flipping
+the default requires dry-run
 evidence.
 
 ### Required evidence before flipping default
@@ -186,7 +186,7 @@ evidence.
    re-evaluate per archetype.
 
 3. **Token contract finalizer passes on every synthesis output.** The
-   finalizer pass (Phase 84 token contract) is the last guardrail
+   finalizer pass (offline-validation token contract) is the last guardrail
    against critic-introduced token drift. If any synthesis output
    fails the finalizer in dry-run, the protocol is not safe to flip.
 

package/skills/create-campaign-v2/core/auto-execute.yaml

@@ -1,4 +1,4 @@
-# Autonomous tail configuration for create-campaign-v2 (Plan 85-02)
+# Autonomous tail configuration for create-campaign-v2
 #
 # This file controls the shell-first review-batch tail that runs AFTER the
 # campaign shell already has the approved source, saved rubrics, and approved
@@ -43,7 +43,7 @@ messaging:
   # relaxed = warn but allow. Default is strict for live campaign state.
   tokenContract: strict
   critique:
-    # Feature flag for the message critique pass shipped in Plan 85-03.
+    # Feature flag for the optional message critique pass.
     # Stays false by default. Flip to true only after the critique
     # protocol has been dry-run against a real sample.
     enabled: false
@@ -82,8 +82,8 @@ messaging:
       # being persisted — synthesis never ships a message that invents
       # tokens, proof, or personalization.
       model: synthesizer
-      # When true, the synthesis step is required to re-run the Phase
-      # 84 finalizer pass on its output. The finalizer is the last
+      # When true, the synthesis step is required to re-run the
+      # offline-validation finalizer pass on its output. The finalizer is the last
       # guardrail against critic-introduced token drift.
       enforceFinalizerPass: true
     opus: