@deftai/directive-content 0.93.0 → 0.94.0

package/packs/strategies/strategies-pack-0.1.json

@@ -121,7 +121,7 @@
         "research"
       ],
       "path": "strategies/research.md",
-      "body": "# Research Strategy\n\nLook before you leap — investigate the domain before planning.\n\nLegend (from RFC2119): !=MUST, ~=SHOULD, ≉=SHOULD NOT, ⊗=MUST NOT, ?=MAY.\n\n**⚠️ See also**: [strategies/interview.md](./interview.md) | [strategies/discuss.md](./discuss.md) | [strategies/map.md](./map.md)\n\n> Adapted from [GSD](https://github.com/gsd-build/get-shit-done) research phase.\n\n---\n\n## When to Use\n\n- ~ Before planning a feature in an unfamiliar domain (auth, payments, real-time, etc.)\n- ~ When the feature involves libraries or APIs the agent hasn't used in this project\n- ? Skip for well-understood domains where the agent has strong existing context\n\n## Output\n\n! Before writing output artifacts, follow the [Preparatory Guard](./artifact-guards.md#preparatory-guard-light).\n\nProduce `vbrief/proposed/{feature}-research.vbrief.json` with two mandatory narratives:\n\n! After emitting this scope vBRIEF, surface the GitHub-issue tracking hint from [emit-hints.md](./emit-hints.md) — name all three patterns (none / `--umbrella` / `--per-vbrief`).\n\n### `DontHandRoll` narrative\n\nProblems that look simple but have existing, battle-tested solutions.\n\n- ! For each problem area, specify: **problem**, **recommended library/tool**, **why not hand-roll**\n- ! Check the project's existing dependencies first -- don't add a library when one is already available\n- ~ Consult official docs for the recommended library (use Context7 or equivalent)\n\n**Example narrative content:**\n```\nProblem: JWT validation\nUse: jose\nRationale: Edge cases in token expiry, key rotation, algorithm confusion\n\nProblem: Email templates\nUse: react-email\nRationale: HTML email rendering is notoriously broken across clients\n\nProblem: Rate limiting\nUse: express-rate-limit\nRationale: IP spoofing, distributed state, Redis integration\n```\n\n### `CommonPitfalls` narrative\n\nWhat goes wrong in this domain, why, and how to avoid it.\n\n- ! For each pitfall: **what happens**, **why it happens**, **how to avoid it**, **warning signs**\n- ~ Informed by library docs, codebase patterns, and known failure modes\n- ~ Prioritize pitfalls that agents specifically tend to hit (stubs, missing error handling, hardcoded values)\n\n**Example narrative content:**\n```\nPitfall: Storing plain-text passwords\nWhat: User passwords saved without hashing\nWhy: Agent implements the happy path and forgets security\nAvoid: Use bcrypt/argon2, never store raw passwords\nWarning signs: No crypto import in auth module, password field stored as-is\n```\n\n### `IPRisk` narrative (#738)\n\n! When the project description, the `Don't Hand-Roll` survey, or the\nresearch notes reference third-party intellectual property (IP), the\nresearch phase MUST run the IP-risk heuristic from\n[`../references/ip-risk.md`](../references/ip-risk.md) -- canonical\nimplementation `scripts/ip_risk.py:detect_ip_terms` -- and persist a\nplain-English `IPRisk` narrative on the research vBRIEF.\n\nThe heuristic is permissive on purpose: recognizable IP names (Magic:\nThe Gathering, Pokemon, etc.), fictional-universe terms (Hogwarts,\nTatooine), branded characters, sports leagues, and trademarked products\nall trigger a hit.\n\n- ! When `detect_ip_terms` returns at least one hit, the research output\n  MUST: (1) ask the explicit monetization-intent question (personal vs\n  commercial); (2) emit `plain_risk_summary(hits, intent)` into the\n  `IPRisk` narrative; (3) plan to inject the protection scope items\n  (`ip_risk_scope_items(intent)`) at SPECIFICATION-generation time.\n- ! On `commercial` intent, surface the **non-optional** lawyer-\n  consultation recommendation in the research output -- this carries\n  forward into the interview output and the SPECIFICATION via the\n  `IPRisk` narrative.\n- ⊗ Treat the absence of detected terms as proof that the project is\n  IP-free. The heuristic only knows about the curated lists; when the\n  research scope is vague, ask the user directly whether the project is\n  based on a game / film / sports league / brand.\n- ⊗ Provide legal advice -- Deft is not a law firm. The only\n  recommendation it makes is **consult a lawyer**.\n\n---\n\n## How Research Feeds Downstream\n\n- ! **Planning** reads research before task decomposition — acceptance criteria account for pitfalls\n- ! **Execution** references \"Don't Hand-Roll\" — agent uses recommended libraries, not custom code\n- ~ **Verification** checks for pitfall warning signs during stub detection\n\n## Research Scope Rules\n\n- ! Research the **current feature only** — not the entire project\n- ! Time-box research — if it takes longer than the feature, scope is wrong\n- ⊗ Research as a reason to delay execution indefinitely\n- ~ Research persists as a vBRIEF in `vbrief/proposed/`\n\n---\n\n## Then: Chaining Gate\n\nAfter research is complete, return to the [chaining gate](./interview.md#chaining-gate)\nso the user can run additional preparatory strategies or proceed to spec generation.\n\n- ! On completion, register artifacts in `./vbrief/plan.vbrief.json`:\n  - Update `completedStrategies`: increment `runCount` for `\"research\"`,\n    append artifact path (`vbrief/proposed/{feature}-research.vbrief.json`)\n  - Append the path to the flat `artifacts` array\n- ! Return to [interview.md Chaining Gate](./interview.md#chaining-gate)\n- ! The research findings MUST inform subsequent strategies and spec generation:\n  - \"Don't Hand-Roll\" items become constraints in the specification\n  - \"Common Pitfalls\" become acceptance criteria or NFRs\n- ⊗ End the session after research without returning to the chaining gate\n  or the invoking strategy's next-step menu\n\n! **Standalone context:** If invoked from a standalone strategy (e.g. map's\n  standalone next-step menu) rather than from the interview chaining gate,\n  return to the invoking strategy's menu instead.\n\n---\n\n## Workflow\n\n1. **Scope** -- Identify the domain and feature boundaries for research\n2. **Survey** -- Check existing project dependencies, official docs, and known pitfalls\n3. **Document** -- Produce `vbrief/proposed/{feature}-research.vbrief.json` with `DontHandRoll` and `CommonPitfalls` narratives\n4. **Chain** -- Return to [interview.md Chaining Gate](./interview.md#chaining-gate), or -- if invoked from a standalone strategy (e.g. map's standalone next-step menu) -- return to the invoking strategy's menu per the [standalone-context rule](#then-chaining-gate) above\n\n## Anti-Patterns\n\n- ⊗ Building custom solutions for solved problems\n- ⊗ Skipping research for unfamiliar domains (\"how hard can auth be?\")\n- ⊗ Research that produces a reading list instead of actionable guidance\n- ⊗ Research that doesn't flow into planning (written and never referenced)\n- ⊗ Ending after research without chaining into specification generation (chained mode; in standalone context, returning to the invoking strategy's menu satisfies the completion requirement per the [standalone-context rule](#then-chaining-gate))\n"
+      "body": "# Research Strategy\n\nLook before you leap — investigate the domain before planning.\n\nLegend (from RFC2119): !=MUST, ~=SHOULD, ≉=SHOULD NOT, ⊗=MUST NOT, ?=MAY.\n\n**⚠️ See also**: [strategies/interview.md](./interview.md) | [strategies/discuss.md](./discuss.md) | [strategies/map.md](./map.md)\n\n> Adapted from [GSD](https://github.com/gsd-build/get-shit-done) research phase.\n\n---\n\n## When to Use\n\n- ~ Before planning a feature in an unfamiliar domain (auth, payments, real-time, etc.)\n- ~ When the feature involves libraries or APIs the agent hasn't used in this project\n- ? Skip for well-understood domains where the agent has strong existing context\n\n## Scope Confirmation Gate (#1273)\n\n! Before autonomous research begins, present one blocking scope-confirmation prompt and wait for the user's selection. Use the deterministic question contract: the final two numbered options MUST be `Discuss` and `Back`.\n\nPrompt:\n> \"What should this research focus on before I investigate autonomously?\"\n\n1. Confirm the inferred feature/domain scope (Recommended)\n2. Refine the feature boundary or priority areas\n3. Provide sample data, artifacts, constraints, or sensitive areas to account for\n4. Discuss\n5. Back\n\n- ! **Confirmed-scope postcondition (required before survey):** research MUST NOT start the survey until a confirmed research scope is recorded in notes. Confirmation is achieved by **any** of: option **1** (accept inferred scope as-is), or option **2**/**3** after free-form capture (the free-form answer **is** the confirmation of scope — it replaces option-1 confirmation; do not re-open option 1 after capture).\n- ! On option **1** (confirm inferred scope): record the inferred feature/domain as the confirmed research scope and proceed to the survey step.\n- ! On option **2** (refine boundary): ask a **follow-up free-form question** in the next message (one question only) to capture the refined feature boundary or priority areas; wait for the user's answer; record that free-form text as the **confirmed** research scope; then proceed to survey. ⊗ Proceed to survey after option 2 without collecting free-form refinement text. ⊗ Leave research blocked after option 2 with no follow-up path.\n- ! On option **3** (artifacts/constraints/sensitivity): ask a **follow-up free-form question** in the next message (one question only) for sample data paths, artifacts to analyze, constraints, or sensitive areas; wait for the user's answer; record provided artifacts/constraints/sensitivity flags **and** treat the current feature/domain (plus those inputs) as the **confirmed** research scope; then proceed to survey. ⊗ Proceed to survey after option 3 without collecting free-form artifact/constraint input. ⊗ Leave research blocked after option 3 with no follow-up path.\n- ! If the user declines free-form input after option 2 or 3 (empty answer / \"skip\" / \"none\"): re-present the Scope Confirmation Gate once; if they pick option **1**, confirm inferred scope and proceed; if they again decline capture without confirming, stop research and return to the chaining gate or invoking menu — do not survey on unconfirmed scope.\n- ! Record the confirmed scope, any provided artifacts, and any sensitivity flags in the research notes before the survey step.\n- ⊗ Start the survey from project description alone without a confirmed research scope (option 1 acceptance or option 2/3 free-form confirmation).\n\n## Output\n\n! Before writing output artifacts, follow the [Preparatory Guard](./artifact-guards.md#preparatory-guard-light).\n\nProduce `vbrief/proposed/{feature}-research.vbrief.json` with two mandatory narratives:\n\n! After emitting this scope vBRIEF, surface the GitHub-issue tracking hint from [emit-hints.md](./emit-hints.md) — name all three patterns (none / `--umbrella` / `--per-vbrief`).\n\n### `DontHandRoll` narrative\n\nProblems that look simple but have existing, battle-tested solutions.\n\n- ! For each problem area, specify: **problem**, **recommended library/tool**, **why not hand-roll**\n- ! Check the project's existing dependencies first -- don't add a library when one is already available\n- ~ Consult official docs for the recommended library (use Context7 or equivalent)\n\n**Example narrative content:**\n```\nProblem: JWT validation\nUse: jose\nRationale: Edge cases in token expiry, key rotation, algorithm confusion\n\nProblem: Email templates\nUse: react-email\nRationale: HTML email rendering is notoriously broken across clients\n\nProblem: Rate limiting\nUse: express-rate-limit\nRationale: IP spoofing, distributed state, Redis integration\n```\n\n### `CommonPitfalls` narrative\n\nWhat goes wrong in this domain, why, and how to avoid it.\n\n- ! For each pitfall: **what happens**, **why it happens**, **how to avoid it**, **warning signs**\n- ~ Informed by library docs, codebase patterns, and known failure modes\n- ~ Prioritize pitfalls that agents specifically tend to hit (stubs, missing error handling, hardcoded values)\n\n**Example narrative content:**\n```\nPitfall: Storing plain-text passwords\nWhat: User passwords saved without hashing\nWhy: Agent implements the happy path and forgets security\nAvoid: Use bcrypt/argon2, never store raw passwords\nWarning signs: No crypto import in auth module, password field stored as-is\n```\n\n### `IPRisk` narrative (#738)\n\n! When the project description, the `Don't Hand-Roll` survey, or the\nresearch notes reference third-party intellectual property (IP), the\nresearch phase MUST run the IP-risk heuristic from\n[`../references/ip-risk.md`](../references/ip-risk.md) -- canonical\nimplementation `scripts/ip_risk.py:detect_ip_terms` -- and persist a\nplain-English `IPRisk` narrative on the research vBRIEF.\n\nThe heuristic is permissive on purpose: recognizable IP names (Magic:\nThe Gathering, Pokemon, etc.), fictional-universe terms (Hogwarts,\nTatooine), branded characters, sports leagues, and trademarked products\nall trigger a hit.\n\n- ! When `detect_ip_terms` returns at least one hit, the research output\n  MUST: (1) ask the explicit monetization-intent question (personal vs\n  commercial); (2) emit `plain_risk_summary(hits, intent)` into the\n  `IPRisk` narrative; (3) plan to inject the protection scope items\n  (`ip_risk_scope_items(intent)`) at SPECIFICATION-generation time.\n- ! On `commercial` intent, surface the **non-optional** lawyer-\n  consultation recommendation in the research output -- this carries\n  forward into the interview output and the SPECIFICATION via the\n  `IPRisk` narrative.\n- ⊗ Treat the absence of detected terms as proof that the project is\n  IP-free. The heuristic only knows about the curated lists; when the\n  research scope is vague, ask the user directly whether the project is\n  based on a game / film / sports league / brand.\n- ⊗ Provide legal advice -- Deft is not a law firm. The only\n  recommendation it makes is **consult a lawyer**.\n\n---\n\n## How Research Feeds Downstream\n\n- ! **Planning** reads research before task decomposition — acceptance criteria account for pitfalls\n- ! **Execution** references \"Don't Hand-Roll\" — agent uses recommended libraries, not custom code\n- ~ **Verification** checks for pitfall warning signs during stub detection\n\n## Research Scope Rules\n\n- ! Research the **current feature only** — not the entire project\n- ! Time-box research — if it takes longer than the feature, scope is wrong\n- ⊗ Research as a reason to delay execution indefinitely\n- ~ Research persists as a vBRIEF in `vbrief/proposed/`\n\n---\n\n## Then: Chaining Gate\n\nAfter research is complete, return to the [chaining gate](./interview.md#chaining-gate)\nso the user can run additional preparatory strategies or proceed to spec generation.\n\n- ! On completion, register artifacts in `./vbrief/plan.vbrief.json`:\n  - Update `completedStrategies`: increment `runCount` for `\"research\"`,\n    append artifact path (`vbrief/proposed/{feature}-research.vbrief.json`)\n  - Append the path to the flat `artifacts` array\n- ! Return to [interview.md Chaining Gate](./interview.md#chaining-gate)\n- ! Present the chaining gate as a blocking question and wait for a user selection before any spec generation or additional scope vBRIEF generation.\n- ! Explain at handoff that `completedStrategies` records that research ran, while `vbrief/proposed/{feature}-research.vbrief.json` remains a planning artifact in the scope lifecycle until a later strategy promotes or consumes it.\n- ! The research findings MUST inform subsequent strategies and spec generation:\n  - \"Don't Hand-Roll\" items become constraints in the specification\n  - \"Common Pitfalls\" become acceptance criteria or NFRs\n- ⊗ Generate implementation scope vBRIEFs directly from research findings or proceed to spec generation before the user chooses from the chaining gate.\n- ⊗ End the session after research without returning to the chaining gate\n  or the invoking strategy's next-step menu\n\n! **Standalone context:** If invoked from a standalone strategy (e.g. map's\n  standalone next-step menu) rather than from the interview chaining gate,\n  return to the invoking strategy's menu instead.\n\n---\n\n## Workflow\n\n1. **Scope confirmation** -- Ask the blocking scope-confirmation prompt, wait for the user, and record scope/artifact/sensitivity inputs\n2. **Survey** -- Check existing project dependencies, official docs, and known pitfalls\n3. **Document** -- Produce `vbrief/proposed/{feature}-research.vbrief.json` with `DontHandRoll` and `CommonPitfalls` narratives\n4. **Chain** -- Return to [interview.md Chaining Gate](./interview.md#chaining-gate), or -- if invoked from a standalone strategy (e.g. map's standalone next-step menu) -- return to the invoking strategy's menu per the [standalone-context rule](#then-chaining-gate) above\n\n## Anti-Patterns\n\n- ⊗ Building custom solutions for solved problems\n- ⊗ Skipping research for unfamiliar domains (\"how hard can auth be?\")\n- ⊗ Starting autonomous research before the Scope Confirmation Gate has captured or explicitly skipped user-provided artifacts/constraints\n- ⊗ Research that produces a reading list instead of actionable guidance\n- ⊗ Research that doesn't flow into planning (written and never referenced)\n- ⊗ Ending after research without chaining into specification generation (chained mode; in standalone context, returning to the invoking strategy's menu satisfies the completion requirement per the [standalone-context rule](#then-chaining-gate))\n"
     },
     {
       "id": "roadmap",

package/skills/deft-directive-review-cycle/SKILL.md

@@ -270,7 +270,7 @@ Remediation:
    - `gh pr checks <N>`
    - `gh api repos/<owner>/<repo>/pulls/<N> -q .head.sha` for HEAD pin
    - `gh api repos/<owner>/<repo>/commits/<sha>/check-runs` for Greptile terminal check-run
-5. Evaluate the same Step 6 fail-closed all-of (terminal check-run + HEAD SHA + Last reviewed commit + confidence > 3 + no P0/P1).
+5. Evaluate the same Step 6 fail-closed all-of (terminal check-run + HEAD SHA + Last reviewed commit + confidence meets resolved min (policy/dogfood/default; see Step 6 #3095) + no P0/P1).
 6. Surface missing-task once to the operator/parent on first detection; do not silently rebrand freestyle sleep as `pr:watch`.
 
 ⊗ Fake a successful `pr:watch` or `review-monitor:*` gate when every dual-invoke probe failed.
@@ -278,7 +278,7 @@ Remediation:
 ⊗ Invent ad-hoc `sleep` / main-session poll / OpenClaw cron loops outside Approach 1–3 when the skill already names this fallback (#2878 / statusreport#153 recurrence).
 ⊗ Skip Step 6 fail-closed fields because deterministic tasks are missing — the gh surfaces above remain mandatory.
 
-! When `clean_gate_holdout=ci_failures` and Greptile otherwise satisfies the probe-side Step 6 fields (SHA match on HEAD, confidence > 3, no P0/P1, not errored): **MUST NOT** idle-poll hoping CI heals. Treat Greptile CLEAN + CI red with the **same ownership** as a Greptile P0 for a merge-ready worker — one fix batch, re-push, re-probe.
+! When `clean_gate_holdout=ci_failures` and Greptile otherwise satisfies the probe-side Step 6 fields (SHA match on HEAD, confidence meets resolved min (policy/dogfood/default; see Step 6 #3095), no P0/P1, not errored): **MUST NOT** idle-poll hoping CI heals. Treat Greptile CLEAN + CI red with the **same ownership** as a Greptile P0 for a merge-ready worker — one fix batch, re-push, re-probe.
 
 ! On persistent `ci_failures` holdout: exit the Greptile wait immediately, fetch failing check annotations (`gh pr checks <N>`, CodeQL / required-check details, or `ci_failed_checks` from `pr:watch --json`), fix or escalate with evidence, then re-enter the review loop after CI is green.
 
@@ -324,6 +324,35 @@ Remediation:
 
 ⊗ Treat a stall as silent -- if the elapsed clock crosses the 10-minute threshold the agent MUST surface the menu, even if the agent is mid-poll. Continuing to poll past the threshold without user input is forbidden.
 
+### Owner Continuity Gate (#3090)
+
+! After any claim of `drive-to: merge-ready`, babysit, shepherd, or "driving review-cycle / merge" on an **open PR**, the **same turn** MUST end in exactly one of:
+
+- **A.** Approach 1 monitor live + sticky `<!-- deft:review-owner -->` lease registered (or #2878 gh-only lease equivalent). Parent yields with an **active** child / monitor and emits `review_cycle: in_progress:<pr>#<monitor_or_lease_ref>`.
+- **B.** Parent-retained ownership: parent does **not** mark the work complete; the next concrete action is an explicit dual-source poll/fix (Step 1 + Step 6). Emit `review_cycle: in_progress:<pr>#parent-retained` (or lease id when a sticky lease exists).
+- **C.** Explicit finish to caller/principal: `BLOCKED` / needs decision / `FAILED` with PR URL + HEAD SHA + why. Emit `review_cycle: skipped:<reason>` or keep `in_progress` only when a live owner remains; never freeform `started`.
+
+! **`review_cycle` evidence enum (portable — all consumers / handoffs / preamble §11 / swarm finish messages):** **only**
+  - `done` — Step 6 fail-closed all-of met on current HEAD (terminal check + HEAD pin + confidence bar + **0 P0/P1** via **dual-source**)
+  - `in_progress:<pr>#<monitor_or_lease_ref>` — verifiable sticky lease, registered monitor id, or documented `parent-retained` ownership
+  - `skipped:<reason>` — intentional skip (e.g. no-pr, operator cancel)
+  - `n/a` — work never entered review-cycle
+
+! Layer **L4** `status: pass` (or equivalent process-green handoff) is **illegal** unless `review_cycle: done` **or** `review_cycle: in_progress:…` with **verifiable** sticky lease / parent-retained ownership documented on that turn.
+
+! Opening a PR may use `in_progress` + lease; the **merge path** requires `done` (stricter product merge bars remain out of scope).
+
+! When dual-source fetch shows open P0/P1 under Step 6 / CLEAN evaluation: continue the fix loop **or** exit **BLOCKED** to the parent — never idle.
+
+! Optional machine gate: `deft verify:l4-owner --pr <N>` (or `task verify:l4-owner -- --pr <N>`; dual-invoke same order as other gates). Exit **0** only when a sticky lease is fresh on the PR **or** the caller asserts `--review-cycle done` after Step 6 (lease-or-done machine gate). `skipped` / `n/a` / parent-retained are process evidence only and do **not** satisfy this machine gate. Exit **1** on silent hold (no lease, no done). Pair with existing monitor-without-lease regression (#2797).
+
+⊗ End an owning turn with **0 children**, **no sticky lease**, and **no finish signal** after a drive-to-merge / babysit / shepherd claim (**silent hold**).
+⊗ Treat check-run **SUCCESS alone** as CLEAN or merge-ready while dual-source P0/P1 remain open.
+⊗ Emit freeform `review_cycle: started` / `pending` / `initiated` or L4 `status: pass` without **A** (or parent-retained **B** with explicit next action) or full Step 6 `done`.
+⊗ Solve Owner Continuity via host cron-as-Approach-1 or always-block-parent-until-merge — use A/B/C above (#2876 / #3090).
+
+~ **Eval / regression (#3090):** Given PR open + check SUCCESS + open inline P1s + agent text claims driving merge + turn ends with 0 subagents and no lease → **FAIL** (Owner Continuity Gate), not PASS.
+
 ### Review Monitoring
 ! **Background / independent dispatch (#1880 Gap D):** Long-running review-cycle owners and pollers (>~3 min) MUST be dispatched independently / in the background so the parent conversation stays interactive. On Cursor, use the Task tool background path (`run_in_background: true`). This generalizes the Approach-1 sub-agent monitor rule to implementation and fix workers as well — foreground dispatch is reserved for short tasks. The parent receives completion via `DONE` / `BLOCKED` / `FAILED` per `templates/agent-prompt-preamble.md` §11.
 
@@ -372,7 +401,7 @@ Remediation:
 
 1. ! Launch via the matching primitive: `start_agent` (Warp), `spawn_subagent` (grok-build / TUI / non-Warp), the Cursor `Task` tool with `run_in_background: true` (`cursor-composer` / `cursor-cloud-agent`, #1877), **or OpenClaw `sessions_spawn` (`openclaw`, #2876)** with a prompt that instructs it to poll for Greptile completion. For `spawn_subagent`, Cursor `Task`, and OpenClaw `sessions_spawn` the prompt MUST reference the canonical poller template `templates/swarm-greptile-poller-prompt.md` (with placeholders filled) plus the agent preamble; the working directory / context must be the PR branch (worktree or equivalent for hybrid).
 2. ! The sub-agent polls using the mechanism for its primitive: for `spawn_subagent` use `get_command_or_subagent_output` (adaptive cadence: ~20-30s first check after push, ~60s second, ~90s thereafter; Greptile typically lands in 3-7 min); for `start_agent` the native messaging path; for the Cursor `Task` tool the backgrounded-task completion-notification path; for OpenClaw `sessions_spawn` the host session completion / messaging channel (prefer `task pr:watch` inside the child when available). Front-load the first check to catch fast reviews.
-3. ! When the exit condition is met (Greptile review current on the HEAD commit SHA, confidence > 3, no P0/P1 remaining), the sub-agent reports completion back to the parent (via `send_message_to_agent`, the spawn_subagent result channel, or the OpenClaw sessions completion channel).
+3. ! When the exit condition is met (Greptile review current on the HEAD commit SHA, confidence meets resolved min (policy/dogfood/default; see Step 6 #3095), no P0/P1 remaining), the sub-agent reports completion back to the parent (via `send_message_to_agent`, the spawn_subagent result channel, or the OpenClaw sessions completion channel).
 4. ! The main conversation pane stays fully interactive during monitoring -- the user (or parent monitor) can continue other work.
 5. ! On receiving the completion message / result, the parent re-fetches findings (both gh pr view --comments and the secondary source) and proceeds to Step 5.
 
@@ -389,6 +418,11 @@ Remediation:
 
 ~ Recurrence: enterprize PR #43 (2026-08-02) — first monitor polled live, host settled empty/unknown; parent spawned a second same-`taskName` monitor; dual lease collision while PR stayed open. See also `meta/lessons.md` and FC04 / growth friction R1 + R10.
 
+
+### Completion latch on settle replay (#3092)
+
+! **Empty settle** is not a latch hit — use Empty announce ≠ done (#3044) above. When the parent has already emitted **one** consolidate for a monitor/leaf `runId` and the harness re-delivers an **identical or equivalent** non-empty settle for the same key, stay **silent** (no re-QC, no second final; host silent token when defined). Re-open only on new runId, principal reopen, or material new evidence. Full MUST: `templates/agent-prompt-preamble.md` §11.5.
+
 ### Single review-monitor lease (#3044 / #2814)
 
 ! **One sticky lease per PR:** ownership is the single sticky GitHub PR comment `<!-- deft:review-owner -->` (or the dual-invoke `review-monitor:register` form that writes it). Parallel ownership is forbidden.
@@ -452,7 +486,7 @@ NOTES: <short>
    - Second check: wait ~60 seconds, then poll
    - Subsequent checks: wait ~90 seconds, then poll
 2. ! Poll using `gh pr view <number> --comments` and `gh pr checks <number>` in the same shell session
-3. ! When the exit condition is met (Greptile review current, confidence > 3, no P0/P1), exit the loop and proceed to Step 5
+3. ! When the exit condition is met (Greptile review current, confidence meets resolved min (policy/dogfood/default; see Step 6 #3095), no P0/P1), exit the loop and proceed to Step 5
 4. ! If the user interrupts (Ctrl+C or equivalent), exit gracefully and report current review status
 
 ! Greptile may advance its review by **editing an existing PR issue comment** rather than creating a new PR review object. Do NOT rely solely on `pulls/{number}/reviews` — that endpoint may remain stale at an older commit SHA even after Greptile has reviewed the latest commit.
@@ -493,7 +527,7 @@ NOTES: <short>
     ```
 
    A missing or non-matching completion marker is `unknown`. See [`../../templates/swarm-greptile-poller-prompt.md`](../../templates/swarm-greptile-poller-prompt.md) `### Last reviewed commit:` for the canonical regex shared with the push-driven poller loop.
-4. ! **Confidence > 3** — the parsed `Confidence Score: X/5` is strictly greater than 3 (i.e. 4/5 or 5/5). A `confidence == 3`, an unparsed confidence, or an absent confidence is `unknown`.
+4. ! **Confidence meets resolved min (#3095)** — the parsed `Confidence Score: X/5` is **>=** the resolved floor from `plan.policy.review.minGreptileConfidence` (inspect: `task policy:show --field=minGreptileConfidence`; `pr:watch` / `pr:merge-ready` share this SoT). Resolution order: typed project policy > framework dogfood detect (framework source → **5**) > consumer default (**4**, legacy confidence > 3 / 4/5+). A score below the floor, an unparsed confidence, or an absent confidence is `unknown`. Directive dogfood MUST NOT exit CLEAN on 4/5.
 5. ! **No P0/P1 findings** — the triple-tier (+ Tier 2.5) detector reports zero P0 and zero P1 findings (P2 issues are non-blocking style suggestions and do not gate the loop).
 
 ! All five fields MUST hold on the SAME single fresh fetch. The agent MUST NOT assemble a "pass" by combining a terminal check-run observed on one poll with a confidence parsed on an earlier poll — the read is atomic per the SHA-pinned-AT-READ-TIME rule above.
@@ -585,6 +619,9 @@ task lifecycle:event -- emit plan:approved \
 
 ## Anti-Patterns
 
+- ⊗ End owning turn with 0 children, no sticky lease, and no finish after drive-to-merge / babysit / shepherd claim — silent hold (#3090)
+- ⊗ Emit freeform `review_cycle: started` / `pending` / `initiated` or L4 `status: pass` without `done` or verifiable `in_progress:<pr>#…` lease/parent-retained (#3090)
+- ⊗ Treat check-run SUCCESS alone as CLEAN / merge-ready while dual-source P0/P1 remain (#3090)
 - ⊗ Ignore [`coding/review.md`](../../coding/review.md) while running this adapter — universal batch/severity/exit/post-merge rules live there
 - ⊗ Route PR shepherding to Cursor global `babysit` on Deft-managed repos when `.deft/core/` is installed -- use this review-cycle skill instead (#2261)
 - ⊗ Route OpenClaw babysit/shepherd/watch to main-session gh poll + cron when `sessions_spawn` is available -- use Approach 1 with `sessions_spawn` (#2876 / #2261)

package/skills/deft-directive-setup/SKILL.md

@@ -655,7 +655,9 @@ omit = [
 - **anything else** (discuss, yolo, speckit, research, brownfield, map, etc.) →
   1. ! Read `deft/strategies/{strategy-name}.md` **right now, in this same turn**
   2. ! Begin the strategy's workflow immediately — ask its first question
-  3. ! **STOP reading this section** — do NOT use the interview process below
+  3. ! For `research`, the first question is the strategy's Scope Confirmation Gate (#1273); ask it and wait before any autonomous research begins.
+  4. ! For `research`, after the research artifact is written, surface the strategy's Then: Chaining Gate and wait for a user selection; do NOT create scope xBRIEFs from research output unless the user later chooses a spec-generating path; do NOT fall through to the interview output path after research.
+  5. ! **STOP reading this section** — do NOT use the interview process below
 
 - ⊗ Default to interview without reading PROJECT-DEFINITION.xbrief.json
 - ⊗ Continue reading below when PROJECT-DEFINITION.xbrief.json specifies a non-interview strategy
@@ -760,8 +762,14 @@ Per [strategies/interview.md](../../strategies/interview.md#interview-rules-shar
 **Spec Structure (both paths):**
 - ! Overview, Architecture
 - ! Implementation Plan: scope xBRIEFs in `xbrief/proposed/` with phases and dependencies
-- ! Explicit dependency mapping between scopes (via xBRIEF `edges` or `references`)
+- ! Explicit dependency mapping MUST use the field consumers actually read for the scope shape:
+  - **Story-shaped scopes** (`plan.metadata.kind = "story"` or scopes intended for swarm allocation / decompose): sequential/blocked work MUST set `plan.metadata.swarm.depends_on` to an array of **resolvable story identifiers**. Swarm readiness, decompose, and queue traversal read **only** this field for story ordering — not `plan.metadata.dependencies` alone. ! Each `depends_on` entry MUST equal the blocking scope's `plan.id` when that field is set; otherwise the blocking artifact's **filename stem** (basename with `.xbrief.json` / `.vbrief.json` stripped — the same stem readiness uses as `story_id`). ⊗ Use rendered titles, free-form prose, date-only prefixes, or unstripped full filenames — those do not resolve and leave the generated scope blocked.
+  - **Phase/epic or cross-scope roadmap batches**: MAY also set plan-level `plan.metadata.dependencies` for roadmap/export readers.
+  - `edges` / `references` may supplement documentation but **do not** replace `plan.metadata.swarm.depends_on` for sequential story scopes.
+- ! When multiple scopes are produced in one Phase 3 pass, encode machine-readable dependency ordering before finishing the write: independent scopes use empty `plan.metadata.swarm.depends_on` (`[]`); sequential/blocked story scopes use non-empty `plan.metadata.swarm.depends_on`. Optionally mirror the same DAG in `plan.metadata.dependencies` for roadmap views.
 - ~ Scopes designed for parallel work by multiple agents
+- ⊗ Deposit sequential story-shaped scopes with only `plan.metadata.dependencies` / `edges` / `references` and missing `plan.metadata.swarm.depends_on` — orchestration will treat them as independent or reject readiness.
+- ⊗ Deposit multiple generated scope xBRIEFs with no dependency metadata and rely on filenames or human prose for ordering.
 - ! Testing Strategy and Deployment captured in narratives
 - ⊗ Write code — specification only
 

package/skills/deft-directive-swarm/SKILL.md

@@ -115,6 +115,12 @@ Large multi-host skills use a **host-neutral core** plus **one** per-host adapte
 ⊗ Multi-sentence progress-only first response after announce with zero tools / yield (#2943 text-repetition hang).
 ⊗ Treat thin DONE (no PR URL / merge evidence) as success (#2943).
 
+
+
+### Completion latch after first consolidate (#3092)
+
+! After the parent emits **one** user/caller-visible consolidate for a child `runId` / settle batch, **identical or equivalent completion replay** for the same key MUST be **silent** (no tools, no re-QC, no second final; host silent token when defined, e.g. OpenClaw `NO_REPLY`). Re-open only on new `runId`/batch, principal explicit reopen, or materially new evidence (new HEAD, new blocker class). Replay storms: at most one fail-loud note, then silent. Full MUST: `templates/agent-prompt-preamble.md` §11.5. Depth: `references/core-phase-5-6.md`. Orthogonal to empty settle ≠ done (#3044).
+
 ## Runtime Capability Detection (summary)
 
 ! Before selecting a launch method, probe the environment. Full probe text: [`references/core-phase-3.md`](references/core-phase-3.md).
@@ -152,6 +158,7 @@ Large multi-host skills use a **host-neutral core** plus **one** per-host adapte
 - ⊗ Prose-only phase handoff after cohort complete (“I will spawn…”) (#2934)
 - ⊗ Multi-sentence progress-only first response after leaf announce with zero tools / yield (#2943)
 - ⊗ Treat thin DONE (no PR URL / merge evidence) as success (#2943)
+- ⊗ Second+ user-visible consolidate for the same child runId without new evidence (#3092)
 - ⊗ Assign overlapping files to multiple agents
 - ⊗ Merge before Greptile exit condition (score > 3, no P0/P1)
 - ⊗ Skip Phase 0 approval before Phase 1

package/skills/deft-directive-swarm/references/core-ops.md

@@ -143,3 +143,5 @@ CONSTRAINTS:
 - ⊗ End a cohort phase-boundary turn with only narrative “I will spawn…” / “review next” and zero next-phase tool calls and no explicit terminal status (`blocked` / `awaiting-human` / `done`) (#2934)
 - ⊗ Multi-sentence progress-only first response after leaf completion announce (`subagent_announce` / parent-push) with zero tools / yield — tool-first ground-truth batch or host yield only (#2943 text-repetition hang)
 - ⊗ Treat thin DONE (completion without PR URL / merge evidence) as success — re-dispatch or take over after ground truth (#2943)
+- ⊗ Second+ user-visible consolidate / final for the same child `runId` / settle batch without new evidence or principal reopen — completion latch silent-replay path (`templates/agent-prompt-preamble.md` §11.5 / #3092)
+- ⊗ Full dual-source re-QC solely because the harness re-delivered the same settle event (#3092)

package/skills/deft-directive-swarm/references/core-phase-5-6.md

@@ -37,6 +37,8 @@ For each agent's PR:
 
 ! **Parent tool-first after leaf announce (#2943):** On the first parent turn after a leaf completion event (`subagent_announce` / parent-push / host completion notify), the parent MUST either (a) emit a **tool-first** ground-truth batch (`gh` / `git` / worktree or file status), or (b) **yield** (`sessions_yield` on OpenClaw, or host equivalent). ⊗ Multi-sentence progress-only first response with zero tools / yield — the text-repetition hang class.
 
+! **Completion latch after first consolidate (#3092):** After the parent has emitted **one** user/caller-visible consolidate for a child `runId` / settle batch (accept or reject), **identical or equivalent completion replay** for the same key MUST be **silent** (no tools, no re-QC, no second final; host silent token when defined, e.g. OpenClaw `NO_REPLY`). Re-open only on new `runId`/batch, principal explicit reopen, or **materially new** evidence (new HEAD, new blocker class) — not fat re-embeds of the same rollup. Replay storms: at most one fail-loud note, then silent. Full MUST + anti-patterns: `templates/agent-prompt-preamble.md` §11.5. Orthogonal to empty settle ≠ done (#3044).
+
 ### Complete xBRIEFs
 
 ! The cohort's story xBRIEFs are completed by the deterministic **cohort completion sweep** in Phase 6 (`task swarm:complete-cohort`, Phase 6 Step 1.5 below), which runs AFTER the merge cascade. Do NOT move story xBRIEFs out of `xbrief/active/` before their PRs merge — a pre-merge move creates premature state if the merge cascade fails. This section is where the monitor records, per story, what the post-merge sweep will finalize:

package/strategies/research.md

@@ -22,6 +22,27 @@ Legend (from RFC2119): !=MUST, ~=SHOULD, ≉=SHOULD NOT, ⊗=MUST NOT, ?=MAY.
 - ~ When the feature involves libraries or APIs the agent hasn't used in this project
 - ? Skip for well-understood domains where the agent has strong existing context
 
+## Scope Confirmation Gate (#1273)
+
+! Before autonomous research begins, present one blocking scope-confirmation prompt and wait for the user's selection. Use the deterministic question contract: the final two numbered options MUST be `Discuss` and `Back`.
+
+Prompt:
+> "What should this research focus on before I investigate autonomously?"
+
+1. Confirm the inferred feature/domain scope (Recommended)
+2. Refine the feature boundary or priority areas
+3. Provide sample data, artifacts, constraints, or sensitive areas to account for
+4. Discuss
+5. Back
+
+- ! **Confirmed-scope postcondition (required before survey):** research MUST NOT start the survey until a confirmed research scope is recorded in notes. Confirmation is achieved by **any** of: option **1** (accept inferred scope as-is), or option **2**/**3** after free-form capture (the free-form answer **is** the confirmation of scope — it replaces option-1 confirmation; do not re-open option 1 after capture).
+- ! On option **1** (confirm inferred scope): record the inferred feature/domain as the confirmed research scope and proceed to the survey step.
+- ! On option **2** (refine boundary): ask a **follow-up free-form question** in the next message (one question only) to capture the refined feature boundary or priority areas; wait for the user's answer; record that free-form text as the **confirmed** research scope; then proceed to survey. ⊗ Proceed to survey after option 2 without collecting free-form refinement text. ⊗ Leave research blocked after option 2 with no follow-up path.
+- ! On option **3** (artifacts/constraints/sensitivity): ask a **follow-up free-form question** in the next message (one question only) for sample data paths, artifacts to analyze, constraints, or sensitive areas; wait for the user's answer; record provided artifacts/constraints/sensitivity flags **and** treat the current feature/domain (plus those inputs) as the **confirmed** research scope; then proceed to survey. ⊗ Proceed to survey after option 3 without collecting free-form artifact/constraint input. ⊗ Leave research blocked after option 3 with no follow-up path.
+- ! If the user declines free-form input after option 2 or 3 (empty answer / "skip" / "none"): re-present the Scope Confirmation Gate once; if they pick option **1**, confirm inferred scope and proceed; if they again decline capture without confirming, stop research and return to the chaining gate or invoking menu — do not survey on unconfirmed scope.
+- ! Record the confirmed scope, any provided artifacts, and any sensitivity flags in the research notes before the survey step.
+- ⊗ Start the survey from project description alone without a confirmed research scope (option 1 acceptance or option 2/3 free-form confirmation).
+
 ## Output
 
 ! Before writing output artifacts, follow the [Preparatory Guard](./artifact-guards.md#preparatory-guard-light).
@@ -127,9 +148,12 @@ so the user can run additional preparatory strategies or proceed to spec generat
     append artifact path (`vbrief/proposed/{feature}-research.vbrief.json`)
   - Append the path to the flat `artifacts` array
 - ! Return to [interview.md Chaining Gate](./interview.md#chaining-gate)
+- ! Present the chaining gate as a blocking question and wait for a user selection before any spec generation or additional scope vBRIEF generation.
+- ! Explain at handoff that `completedStrategies` records that research ran, while `vbrief/proposed/{feature}-research.vbrief.json` remains a planning artifact in the scope lifecycle until a later strategy promotes or consumes it.
 - ! The research findings MUST inform subsequent strategies and spec generation:
   - "Don't Hand-Roll" items become constraints in the specification
   - "Common Pitfalls" become acceptance criteria or NFRs
+- ⊗ Generate implementation scope vBRIEFs directly from research findings or proceed to spec generation before the user chooses from the chaining gate.
 - ⊗ End the session after research without returning to the chaining gate
   or the invoking strategy's next-step menu
 
@@ -141,7 +165,7 @@ so the user can run additional preparatory strategies or proceed to spec generat
 
 ## Workflow
 
-1. **Scope** -- Identify the domain and feature boundaries for research
+1. **Scope confirmation** -- Ask the blocking scope-confirmation prompt, wait for the user, and record scope/artifact/sensitivity inputs
 2. **Survey** -- Check existing project dependencies, official docs, and known pitfalls
 3. **Document** -- Produce `vbrief/proposed/{feature}-research.vbrief.json` with `DontHandRoll` and `CommonPitfalls` narratives
 4. **Chain** -- Return to [interview.md Chaining Gate](./interview.md#chaining-gate), or -- if invoked from a standalone strategy (e.g. map's standalone next-step menu) -- return to the invoking strategy's menu per the [standalone-context rule](#then-chaining-gate) above
@@ -150,6 +174,7 @@ so the user can run additional preparatory strategies or proceed to spec generat
 
 - ⊗ Building custom solutions for solved problems
 - ⊗ Skipping research for unfamiliar domains ("how hard can auth be?")
+- ⊗ Starting autonomous research before the Scope Confirmation Gate has captured or explicitly skipped user-provided artifacts/constraints
 - ⊗ Research that produces a reading list instead of actionable guidance
 - ⊗ Research that doesn't flow into planning (written and never referenced)
 - ⊗ Ending after research without chaining into specification generation (chained mode; in standalone context, returning to the invoking strategy's menu satisfies the completion requirement per the [standalone-context rule](#then-chaining-gate))

package/tasks/verify.yml

@@ -496,6 +496,16 @@ tasks:
         vars:
           ENGINE_CMD: 'verify-review-monitor {{.CLI_ARGS}}'
 
+  l4-owner:
+    desc: "Owner Continuity / L4 owner gate (#3090): exit 0 only if sticky review-owner lease is fresh or --review-cycle done (forbid freeform started/pending). Three-state exit (0 ready / 1 silent hold / 2 config)."
+    dir: '{{.USER_WORKING_DIR}}'
+    deps:
+      - task: :engine:_ts-build
+    cmds:
+      - task: :engine:invoke
+        vars:
+          ENGINE_CMD: 'verify-l4-owner {{.CLI_ARGS}}'
+
   subagent-alive:
     desc: "Fail-closed worker liveness gate (#2824): missing/STALE heartbeats for in-flight drive-to:merge* workers exit 1 with REDISPATCH_OK. Three-state exit (0 alive / 1 stale-or-missing / 2 config error)."
     dir: '{{.USER_WORKING_DIR}}'

package/templates/agent-prompt-preamble.md

@@ -545,13 +545,46 @@ Every worker MUST send a final status message before exiting its tool loop, rega
 
 ! **Empty announce ≠ done / single review-monitor lease (#3044 / FC04 residual):** An empty settle, missing `STATUS:` line, or `status: unknown` from a review-monitor (`subagent_announce` with `(no output)` included) is **not** DONE/CLEAN/merge-ready. Parent MUST same-turn ground truth (`gh pr view` + checks + HEAD) and MUST NOT spawn a second monitor while the prior owner is running or only falsely settled. Prefer one sticky `<!-- deft:review-owner -->` lease and a non-empty `STATUS`/`HEAD`/`CHECKS`/`MERGE` handback. Full MUST language: `skills/deft-directive-review-cycle/SKILL.md` + OpenClaw host adapter residual.
 
+! **`review_cycle` evidence enum (#3090):** Handoffs, swarm finish messages, and L4 process claims MUST use only `done` | `in_progress:<pr>#<monitor_or_lease_ref>` | `skipped:<reason>` | `n/a`. Freeform `started` / `pending` / `initiated` is forbidden. L4 `status: pass` is illegal unless `review_cycle: done` (Step 6 fail-closed on HEAD) or `review_cycle: in_progress:…` with a verifiable sticky lease / parent-retained ownership. After a drive-to:merge-ready / babysit / shepherd claim, the same turn MUST end in Owner Continuity Gate A/B/C (monitor+lease, parent-retained next dual-source action, or explicit BLOCKED/FAILED finish) — never silent hold. Optional machine gate: `deft verify:l4-owner --pr <N>` / `task verify:l4-owner -- --pr <N>`. Full MUST language: `skills/deft-directive-review-cycle/SKILL.md` Owner Continuity Gate.
+
 ! **Parent tool-first after leaf completion (#2943):** When a parent / monitor receives a leaf completion event (`subagent_announce`, parent-push, or host completion notify), its **first response** MUST be a **tool-first** ground-truth batch (`gh` / `git` / worktree or file status) **or** a host **yield** (`sessions_yield` on OpenClaw, or equivalent). ⊗ Multi-sentence progress-only first response with zero tools / yield — the OpenClaw text-repetition hang class (#2943).
 
 ⊗ Treat thin DONE (no PR URL / merge evidence) as success (#2943).
 ⊗ Treat empty/unknown review-monitor settle as DONE without same-turn ground truth, or dual-spawn a second monitor while the first lease is live (#3044).
+⊗ Emit freeform `review_cycle: started` / `pending` / `initiated` or L4 `status: pass` without `done` or verifiable `in_progress:<pr>#…` (#3090).
 
 Per-step acks during the run are noise. ONE start message, ONE final message; intermediate messages only on `BLOCKED` / `FAILED`. The final message lets the dispatcher distinguish a clean exit from a silent timeout when the lifecycle event arrives.
 
+## 11.5 Completion latch — one consolidate per runId (#3092)
+
+Multi-agent **orchestrators** (OpenClaw parent seats, Cursor Task parents, grok-build swarm monitors, any parent that receives child settle / completion events) MUST apply a portable **completion latch**. Host-level announce dedupe is complementary and imperfect; this is the **agent-side** default when the same settled batch is re-delivered.
+
+### Completion latch (MUST)
+
+1. **One user- or caller-visible consolidate per child `runId` / settle batch** (or the explicit equivalent batch key the harness provides — e.g. spawn id, task id, announce id). Accept or reject for that batch still counts as the one consolidate.
+2. After that consolidate is emitted, **identical or equivalent completion replay** for the same key ⇒ **silent**: no tools, no re-QC, no second final answer. When the host defines a silent token (example: OpenClaw `NO_REPLY`), use it; otherwise emit no outbound user/caller message.
+3. **Re-open only when:**
+   - new `runId` / new child batch key, **or**
+   - caller / principal **explicit** steer to reopen, **or**
+   - the completion payload carries **materially new** evidence (new HEAD, new blocker class, new PR URL / merge state) — not a re-paste or fat re-embedding of the same rollup / full task text.
+4. If the harness **storms** replays with no new key: **at most one** fail-loud note to the caller (`completion replay storm; ignoring`), then silent. ⊗ Infinite consolidate loops.
+5. Fat completion payloads that re-embed full task text, prior prompts, or prior consolidate prose **MUST NOT** be treated as a new mission or as material new evidence.
+
+### Eval checklist (second settle same runId)
+
+Given: parent already emitted a consolidate for `runId=R` (or harness batch key `R`).  
+Second settle event for `R` arrives with the same claims / equivalent rollup.  
+**Expect:** silent / host silent-token path — **not** a new investigation narrative, dual-source re-fetch, or second user-visible final.
+
+### Normative anti-patterns
+
+- ⊗ Second+ user-visible "final" for the same settled `runId` without new evidence or explicit reopen
+- ⊗ Full dual-source / full test re-run solely because the settle event was delivered again
+- ⊗ Treating "send consolidated final **now**" wording on a **replay** as authorization to undo a prior consolidate for that batch
+- ⊗ Treating fat prompt / task-text re-embeds in completion payloads as a new mission (#3092)
+
+Cross-links: swarm Phase 5 completion-notification / parent handback (`skills/deft-directive-swarm/references/core-phase-5-6.md`); review-monitor empty-settle DoD remains #3044 (empty ≠ done) and is orthogonal — empty is not a latch hit; identical non-empty replay after a prior consolidate **is** a latch hit.
+
 ## 12. Session ritual + `task verify:cache-fresh` gates before `start_agent` (#1348 / #1127)
 
 Dispatchers (this orchestrator, swarm Phase 4 dispatch, monitor agents, scheduled / cloud runs) run in a headless worker context and MUST set `DEFT_SESSION_RITUAL_SKIP=1` for dispatched implementation workers. The interactive parent session remains responsible for `task session:start`; worker processes bypass the local `.deft/ritual-state.json` gate explicitly so they do not need per-clone interactive ritual state. When the bypass would hide a stale/missing ritual state, `task verify:session-ritual` prints a warning to stderr; preserve that warning in the dispatch log.

package/templates/swarm-greptile-poller-prompt.md

@@ -280,7 +280,7 @@ if m is None:
 confidence = int(m.group(1)) if m else None
 ```
 
-The clean threshold is `confidence > 3`, i.e. 4/5 or 5/5. Lower scores indicate Greptile is uncertain -- do NOT exit clean.
+The clean threshold is `confidence >= min_confidence`, where `min_confidence` is the resolved floor from `plan.policy.review.minGreptileConfidence` (#3095): typed project policy > framework dogfood detect (framework source → **5**) > consumer default (**4**, legacy `confidence > 3` / 4/5+). Inspect with `task policy:show --field=minGreptileConfidence`. Lower scores indicate Greptile is uncertain -- do NOT exit clean. Directive dogfood MUST NOT exit CLEAN on 4/5.
 
 ### Informal-clean missing canonical fields (#1543)
 
@@ -333,6 +333,7 @@ def evaluate_clean_gate(
     ci_failures,
     errored,
     terminal_check_run,
+    min_confidence=4,
 ):
     """Return (is_clean, clean_gate_holdout) per the (6)-condition AND gate.
 
@@ -347,21 +348,23 @@ def evaluate_clean_gate(
     non-terminal Greptile conclusion (`queued` / `in_progress` /
     `cancelled` / `timed_out` / `stale` / `action_required` / `failure`)
     is NOT clean even when the rolling summary already parses clean (SHA
-    matches HEAD, confidence > 3, no P0/P1). This is the INCOMPLETE_BUT_RATED
-    scenario from `skills/deft-directive-review-cycle/SKILL.md` Step 6:
-    without (6), all five legacy conditions pass and the poller exits CLEAN
-    prematurely. `ci_failures` (condition 4) is scoped to `CI / *` checks
-    ONLY -- it does NOT cover the Greptile Review check-run -- so the
-    terminal Greptile conclusion is a DISTINCT condition that `ci_failures`
-    cannot stand in for. Mirrors the SKILL.md Step 6 fail-closed all-of's
-    terminal-check-run field so the swarm-dispatched poller path enforces
-    the same gate as the one-shot review-cycle entry.
+    matches HEAD, confidence >= min_confidence, no P0/P1). This is the
+    INCOMPLETE_BUT_RATED scenario from
+    `skills/deft-directive-review-cycle/SKILL.md` Step 6: without (6), all
+    five legacy conditions pass and the poller exits CLEAN prematurely.
+    `ci_failures` (condition 4) is scoped to `CI / *` checks ONLY -- it does
+    NOT cover the Greptile Review check-run -- so the terminal Greptile
+    conclusion is a DISTINCT condition that `ci_failures` cannot stand in
+    for. Mirrors the SKILL.md Step 6 fail-closed all-of's terminal-check-run
+    field so the swarm-dispatched poller path enforces the same gate as the
+    one-shot review-cycle entry. `min_confidence` defaults to the consumer
+    bar (4); resolve via project policy / dogfood (#3095).
     """
     if last_reviewed_sha is None or last_reviewed_sha != head_sha:
         return False, "sha_match"
     if has_blocking:
         return False, "has_blocking"
-    if confidence is None or confidence <= 3:
+    if confidence is None or confidence < min_confidence:
         return False, "confidence"
     if ci_failures > 0:
         return False, "ci_failures"
@@ -417,10 +420,10 @@ When ANY of the six conditions below fires, send the corresponding message to `{
 ALL of:
 - `last_reviewed_sha` parsed and matches the current PR HEAD SHA (compare via `gh pr view {pr_number} --repo {repo} --json headRefOid --jq .headRefOid`).
 - `has_blocking` is False (no P0 / P1 findings).
-- `confidence > 3` (i.e. 4/5 or 5/5 -- a `confidence == 3` parse is NOT clean; the gate names `clean_gate_holdout="confidence"` and you stay in the loop, you do NOT send the CLEAN message).
+- `confidence >= min_confidence` where `min_confidence` is the resolved policy floor (#3095; consumer default 4 / dogfood 5 -- a score below the floor is NOT clean; the gate names `clean_gate_holdout="confidence"` and you stay in the loop, you do NOT send the CLEAN message).
 - `gh pr checks {pr_number}` shows no `failure` status on `CI / *` checks.
 - The Greptile rolling-summary comment body does NOT equal `Greptile encountered an error while reviewing this PR` (errored sentinel; #526).
-- `terminal_check_run` is True: the `Greptile Review` check-run on the current HEAD is terminal -- `status == "completed"` AND `conclusion` in `{{success, neutral}}` (#1259). A non-terminal Greptile conclusion (`queued` / `in_progress` / `cancelled` / `timed_out` / `stale` / `action_required` / `failure`) is NOT clean even when the rolling summary already parses clean (the INCOMPLETE_BUT_RATED scenario -- rolling summary posted, SHA matches, confidence > 3, no P0/P1, but the check-run has not terminally landed). This is DISTINCT from the `CI / *` `failure` bullet above: `ci_failures` is scoped to `CI / *` checks only and does NOT cover the Greptile Review check-run, so a non-terminal Greptile conclusion would otherwise slip through. The gate names `clean_gate_holdout="terminal_check_run"` and you stay in the loop.
+- `terminal_check_run` is True: the `Greptile Review` check-run on the current HEAD is terminal -- `status == "completed"` AND `conclusion` in `{{success, neutral}}` (#1259). A non-terminal Greptile conclusion (`queued` / `in_progress` / `cancelled` / `timed_out` / `stale` / `action_required` / `failure`) is NOT clean even when the rolling summary already parses clean (the INCOMPLETE_BUT_RATED scenario -- rolling summary posted, SHA matches, confidence meets min, no P0/P1, but the check-run has not terminally landed). This is DISTINCT from the `CI / *` `failure` bullet above: `ci_failures` is scoped to `CI / *` checks only and does NOT cover the Greptile Review check-run, so a non-terminal Greptile conclusion would otherwise slip through. The gate names `clean_gate_holdout="terminal_check_run"` and you stay in the loop.
 
 Send to parent:
 
@@ -437,7 +440,7 @@ Send to parent:
       Last reviewed commit: <sha>
       -- no more polling, exiting now
 
-**Swarm-orchestrated terminal contract (#1364):** when this poller is dispatched as part of a swarm cohort (parent monitor is running `skills/deft-directive-swarm/SKILL.md` Phase 6), this exact subject line -- `PR #{pr_number} CLEAN -- ready for merge` -- with `confidence > 3` recorded on the **current HEAD** is the ONLY acceptable "review complete" signal the swarm monitor accepts toward the Phase 5 -> 6 merge-gate transition. The five other terminal exits below ((2) NEW P0/P1 FINDINGS escalation, (3) ERRORED, (4) TIMEOUT, (5) STALL, (6) INFORMAL-CLEAN) are NOT "review complete" signals for swarm purposes: each one MUST force either fresh poller re-dispatch on the same PR or explicit user escalation BEFORE the monitor surfaces the Phase 5 -> 6 gate. The monitor enforces this structurally via `task swarm:verify-review-clean` (#1364); see `skills/deft-directive-swarm/SKILL.md` Phase 5 Exit Condition for the cohort verifier mandate. A poller that has terminated lifecycle-clean (i.e. the sub-agent process exited normally) but with `clean_gate_holdout != None` HAS NOT "reported review-clean" for swarm-cycle purposes -- the verifier picks the gap up and the monitor re-dispatches.
+**Swarm-orchestrated terminal contract (#1364):** when this poller is dispatched as part of a swarm cohort (parent monitor is running `skills/deft-directive-swarm/SKILL.md` Phase 6), this exact subject line -- `PR #{pr_number} CLEAN -- ready for merge` -- with `confidence >= min_confidence` recorded on the **current HEAD** is the ONLY acceptable "review complete" signal the swarm monitor accepts toward the Phase 5 -> 6 merge-gate transition. The five other terminal exits below ((2) NEW P0/P1 FINDINGS escalation, (3) ERRORED, (4) TIMEOUT, (5) STALL, (6) INFORMAL-CLEAN) are NOT "review complete" signals for swarm purposes: each one MUST force either fresh poller re-dispatch on the same PR or explicit user escalation BEFORE the monitor surfaces the Phase 5 -> 6 gate. The monitor enforces this structurally via `task swarm:verify-review-clean` (#1364); see `skills/deft-directive-swarm/SKILL.md` Phase 5 Exit Condition for the cohort verifier mandate. A poller that has terminated lifecycle-clean (i.e. the sub-agent process exited normally) but with `clean_gate_holdout != None` HAS NOT "reported review-clean" for swarm-cycle purposes -- the verifier picks the gap up and the monitor re-dispatches.
 
 ### (2) NEW P0/P1 FINDINGS
 
@@ -604,5 +607,5 @@ Dogfood lessons captured during the #727 self-review cycle. The template body ab
 - `meta/lessons.md` `## Orchestrator Role Separation + Canonical Poller Template (2026-04)` -- short cross-reference; the rule body lives in the skills above (per `main.md` Rule Authority [AXIOM]).
 - #727 -- this template's acceptance issue and the full anti-pattern record (rm-chaining, parsing-bug recurrence, role-conflation in implementation-agent prompts).
 - #1039 -- (5) STALL terminal exit + Tier 1 instrumentation + Tier 3 per-condition fail-loud (`clean_gate_holdout`); the third recurrence in this template's detector-gap chain after #910 (triple-tier) and #1035 (Tier 2.5 + confidence-heading).
-- #1364 -- cohort-level CLEAN verification gate (`task swarm:verify-review-clean`, `scripts/swarm_verify_review_clean.py`). The (1) CLEAN section's swarm-orchestrated terminal contract block declares that only the exact `PR #{pr_number} CLEAN -- ready for merge` subject with `confidence > 3` on current HEAD is an acceptable "review complete" signal for the swarm monitor's Phase 5 -> 6 transition; the cohort verifier picks up any other terminal exit ((2) NEW P0/P1 FINDINGS escalation, (3) ERRORED, (4) TIMEOUT, (5) STALL) and holds the merge gate until fresh poller re-dispatch or explicit user escalation resolves it. Recurrence record: #1166 swarm execution where multiple pollers exited with `clean_gate_holdout=confidence` (confidence == 3) and the monitor still raised the Phase 5 -> 6 gate because the trigger keyed on "all pollers have reported back" rather than "every PR in the cohort is objectively CLEAN".
+- #1364 -- cohort-level CLEAN verification gate (`task swarm:verify-review-clean`, `scripts/swarm_verify_review_clean.py`). The (1) CLEAN section's swarm-orchestrated terminal contract block declares that only the exact `PR #{pr_number} CLEAN -- ready for merge` subject with `confidence >= min_confidence` on current HEAD is an acceptable "review complete" signal for the swarm monitor's Phase 5 -> 6 transition; the cohort verifier picks up any other terminal exit ((2) NEW P0/P1 FINDINGS escalation, (3) ERRORED, (4) TIMEOUT, (5) STALL) and holds the merge gate until fresh poller re-dispatch or explicit user escalation resolves it. Recurrence record: #1166 swarm execution where multiple pollers exited with `clean_gate_holdout=confidence` (confidence == 3) and the monitor still raised the Phase 5 -> 6 gate because the trigger keyed on "all pollers have reported back" rather than "every PR in the cohort is objectively CLEAN". #3095 raises the dogfood floor to 5 via policy.
 - #2879 -- OpenClaw `sessions_spawn` + parent push/announce completion channel named in Role posture (alongside Warp `start_agent` and grok-build `spawn_subagent`); heartbeat mapping in `docs/subagent-heartbeat.md` and preamble §10.5. Epic #2874.

package/tools/greptile.md

@@ -124,7 +124,7 @@ The check run name is `"Greptile Review"` — this must match the context name i
 
 1. Check the `deft-directive-review-cycle` skill's Pre-Flight Check section
 2. Verify `triggerOnUpdates` is enabled — without it, Greptile never produces a follow-up review
-3. Verify the confidence score section is enabled in dashboard settings — the exit condition requires confidence >3
+3. Verify the confidence score section is enabled in dashboard settings — the exit condition requires confidence >= the resolved floor (`plan.policy.review.minGreptileConfidence`; consumer default 4 / legacy >3; directive dogfood 5 — `task policy:show --field=minGreptileConfidence`, #3095)
 
 ## Anti-Patterns