@garygentry/feature-forge 0.2.3 → 0.2.5

package/adapters/codex/skills/forge/SKILL.md

@@ -21,26 +21,26 @@ For pipeline architecture details, read `references/process-overview.md`.
 **If a feature name is provided** (e.g., `/feature-forge:forge auth`):
 - **First test whether the name is an epic:** if `{specsDir}/{name}/epic-manifest.json` exists, render the **Epic Dashboard** (see format below) and stop — do not treat it as a feature.
 - Otherwise, resolve the name via the **Feature Directory Resolution** block in `references/shared-conventions.md` (so a nested epic-member name finds its dashboard too). On a resolution failure (`not-found` / `ambiguous` at exit 1; `unsafe-name` or a path-containment escape at exit 2), surface it verbatim.
-  - On `not-found` for a never-started feature, ask: "No pipeline exists for '{feature}'. Want to start one? Run `/feature-forge:forge-1-prd {feature}` to begin."
+  - On `not-found`, first run the cross-branch discovery step from that same Feature Directory Resolution block (`forge-session.py discover-feature` — the state may live on a topic branch or an unfetched remote branch). Candidates → offer switch / fetch+switch per that block (explicit accept + clean tree only), then re-resolve and render the dashboard. Only when discovery also returns nothing, ask: "No pipeline exists for '{feature}' on any branch. Want to start one? Run `/feature-forge:forge-1-prd {feature}` to begin." Never render a dashboard from memory of earlier sessions (anti-fabrication guard).
 - If resolution succeeds, display the per-feature pipeline status dashboard (see format below) from `{resolvedFeatureDir}/`.
 
 **If no feature name is provided:** list in two tiers.
 
 1. **Epics first.** Identify epic directories as any `{specsDir}/*/` that directly contains an `epic-manifest.json` **and no `.pipeline-state.json` of its own** (an epic root is never itself a feature). For each epic, run:
    ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" render-status "{epic}" --specs-dir "{specsDir}" --json
    ```
    and show one rollup line: `{epic} — {complete}/{total} complete, next: {nextCommand}`.
 2. **Standalone features below.** Scan the remaining `{specsDir}/*/` that directly contain a `.pipeline-state.json` **without** an `epic` back-pointer. A nested member's `.pipeline-state.json` is **attributed to its epic (Tier 1), never listed as a standalone feature**.
-   - **Rank by recency.** Run the recency ranker so the most-recently-touched active feature is the default — the user rarely has to type a name (especially on mobile after a `/clear`):
+   - **Rank by recency.** Run the recency ranker so the most-recently-touched active feature is the default — the user rarely has to type a name (especially on mobile after a clear your session / start a fresh session):
      ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/forge-session.py" rank-features --specs-dir "{specsDir}" --json
      ```
-     This returns `{active: [...], counts: {...}}` with active features sorted by `updatedAt` **descending** (row 0 is the most recent). Each row carries `currentStage`, `nextStage`, `nextCommand`, `verifyPending`, and `verifyCommand` (the single source of stage order). The `active` list excludes nested epic members surfaced in Tier 1 — but the ranker scans them too, so ignore rows whose `epic` is non-null here (they belong to the epic rollup).
+     This returns `{active: [...], counts: {...}}` with active features sorted by `updatedAt` **descending** (row 0 is the most recent). Each row carries `currentStage`, `nextStage`, `nextCommand`, `verifyPending`, `verifyCommand`, `verifyStage`, `verifyState` (`fresh`/`stale`/`failing`/`never`/`none`), `autoVerify` (the effective per-stage setting), and `autoFix` (the single source of stage order). A top-level `invalidAutoVerifyKeys` array appears when `forge.config.json` has `autoVerifyStages` keys outside the five verify-capable stages — surface it as a one-line warning. The `active` list excludes nested epic members surfaced in Tier 1 — but the ranker scans them too, so ignore rows whose `epic` is non-null here (they belong to the epic rollup).
    - **Pick the feature:** if exactly one active standalone pipeline exists, show its dashboard. If multiple exist, use the host's question mechanism — **list the most-recently-updated first, labeled `(recommended)`**, each option's description showing its `currentStage` and a relative age ("updated 2h ago"). Always include a free-form escape ("A different feature / something else") so the user is never boxed in. Then render the chosen feature's dashboard.
 
 If no epics and no standalone features exist, say: "No active feature pipelines found. Start one with `/feature-forge:forge-1-prd <feature-name>` or group several with `/feature-forge:forge-0-epic <epic-name>`."
@@ -86,31 +86,45 @@ Use these status indicators:
 
 ### 3b. Drive to the Next Stage
 
-After rendering a **per-feature** dashboard for an **active** pipeline (skip this for paused/abandoned pipelines and for the Epic Dashboard), don't stop at a text suggestion — actively offer to start the next stage. This removes the copy-paste-after-`/clear` chore that makes long, multi-stage runs painful (especially on mobile).
+After rendering a **per-feature** dashboard for an **active** pipeline (skip this for paused/abandoned pipelines and for the Epic Dashboard), don't stop at a text suggestion — actively offer to start the next stage. This removes the copy-paste-after-clear your session / start a fresh session chore that makes long, multi-stage runs painful (especially on mobile).
 
-**1. Read the next step.** From the `rank-features --json` output (above), find this feature's row and read its `nextStage`, `nextCommand`, `verifyPending`, and `verifyCommand`. If the feature is not in the `active` list (paused/abandoned), or `nextStage` is `null` (every production stage complete), skip the drive prompt — instead congratulate the user and, if `forge-6-docs` has not run, offer it; otherwise note the pipeline is complete.
+**1. Read the next step.** From the `rank-features --json` output (above), find this feature's row and read its `nextStage`, `nextCommand`, `verifyPending`, `verifyCommand`, `verifyStage`, `verifyState`, `autoVerify`, and `autoFix`. If the feature is not in the `active` list (paused/abandoned), or `nextStage` is `null` (every production stage complete), skip the drive prompt — instead congratulate the user and, if `forge-6-docs` has not run, offer it; otherwise note the pipeline is complete. If the payload has a non-empty `invalidAutoVerifyKeys`, print a one-line warning first (e.g. "⚠️ forge.config.json `autoVerifyStages` has unknown keys: … — they are ignored; fix the typo").
 
 **2. Check the context window.** Run the context-usage helper so you can advise whether to continue here or start the next stage in a fresh session:
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/forge-session.py" context-usage --json
 ```
 - `{"available": true, ...}` → note `pct` (e.g. "context ~68% full") and `overThreshold`. Window/threshold come from `contextWindowTokens` / `contextWarnThreshold` in `forge.config.json` (the helper defaults to a 200k window and 0.7 threshold, and auto-bumps the assumed window to 1M once observed usage exceeds 200k; **on a 1M-context model set `contextWindowTokens: 1000000` so the percentage is accurate below 200k too** — 1M can't be detected from the transcript until usage crosses 200k).
 - `{"available": false, ...}` → omit context advice silently (non-Claude host, or a fresh session with no transcript). Never treat this as an error.
 
-**3. Offer the next step via the host's question mechanism.** Output the dashboard + a one-line context note as text, then ask (per the Decision Support protocol in `references/shared-conventions.md`). Options, in this order:
-- **Start `{nextStage}` now** — recommended **when context is healthy** (`overThreshold` false or context unavailable).
-- **Start in a clean session** — recommended-**first** instead **when `overThreshold` is true**. The work survives a clear because all state is on disk: instruct the user to `/clear`, then re-run `/feature-forge:forge {feature}` (or run `{nextCommand}` directly) in the fresh session. Note plainly that you cannot `/clear` for them.
-- **Verify `{stage}` first** — include **only when `verifyPending` is true**; selecting it runs `{verifyCommand}`.
+**2b. Auto-verify catch-up (when `verifyPending` is true).** Under this behavior the just-completed authoring stage runs auto-verify **in-stage** (`references/stage-exit-protocol.md`, in-stage verify block), which clears `verifyPending` — so on the normal path this branch does **nothing**. It fires only as a **catch-up**: `verifyPending` is still true because the producing stage could not dispatch a clean-room subagent (non-Claude host), or ran before this behavior landed. When it does fire, decide whether verify runs automatically (identical logic to the in-stage run, so a stage that already verified is never double-verified):
+
+- **`autoVerify` is true for the just-completed `verifyStage`** → **skip the verify question entirely** and run verify now, *provided it can run clean-room*. Auto-verify is safe to run unattended only because verify executes in a fresh `forge-verifier` subagent that inherits none of this session's context (so no clear your session / start a fresh session is needed and only a compact digest returns). Guard the clean-room assumption: proceed unattended **only when the host's subagent mechanism + `forge-verifier` subagent are available**. Invoke `feature-forge:forge-verify` via the host's skill-invocation mechanism in **require-clean (`auto`) mode** — in that mode forge-verify refuses to run inline and returns a sentinel if the subagent is not dispatchable (see `skills/forge-verify/SKILL.md`). Then:
+  - **Sentinel returned (clean-room unavailable)** → do **not** run verify inline. Degrade to the manual gate: fall through to step 3 with the **"Verify `{stage}` first (manual)"** option included, and if `overThreshold`, recommend "clear your session / start a fresh session, then verify in a clean session." Verify state stays outstanding; the stage is never advanced on false assurance.
+  - **Verify passed / no findings** → proceed to the normal advance gate (step 3), no verify option.
+  - **Verify found findings** →
+    - **`autoFix` is true AND preconditions hold** — the findings document has **zero unresolved decision points** and the **working tree is clean** → invoke `feature-forge:forge-fix` via the host's skill-invocation mechanism, then run a **mandatory re-verify** (require-clean mode). Advance only if the re-verify passes. On **any** precondition miss (decisions required, dirty tree), a forge-fix early stop, or a red re-verify → fall back to the digest + prompt below (never a silent partial mutation).
+    - **`autoFix` is false (default), or preconditions failed** → present a **compact findings digest** as text, then the host's question mechanism: *Apply fixes now (forge-fix)* / *Review findings* / *Skip & advance*.
+- **`autoVerify` is false/unconfigured** → do not auto-run; use the advance gate in step 3, which gains the folded opt-in verify options ("Verify `{stage}` now" and "Verify `{stage}` now + enable auto-verify going forward").
+
+**3. Offer the next step via the host's question mechanism.** (Reached when auto-verify did not fully resolve the step, or `verifyPending` is false.) Output the dashboard + a one-line context note as text, then ask (per the Decision Support protocol in `references/shared-conventions.md`). This is the same **Stage Exit Protocol** the stage skills stamp (`references/stage-exit-protocol.md`) — a clean session at the boundary is recommended on its own merits, and `overThreshold` only modulates *how emphatically*, never *whether*. Options, in this order:
+- **Start `{nextStage}` in a clean session** — **recommended, unconditionally**, at every boundary. A clean session is the right default for the next stage on its own merits, independent of window size. The work survives a clear because all state is on disk: instruct the user to clear your session / start a fresh session, then re-run `/feature-forge:forge {feature}` (or run `{nextCommand}` directly) in the fresh session. Note plainly that you cannot clear your session / start a fresh session for them. **When `overThreshold` is true**, strengthen this wording (the window is also genuinely full) and add mid-stage compaction advice; **when the window is healthy**, still recommend the clean session — `overThreshold` is a *secondary, additive* modifier here, not the clear on/off switch.
+- **Continue in this session** — the always-available alternative, reasonable when the window is nearly empty and the user would rather not clear. Not the default.
+- **Verify `{stage}` now** — include **only when `verifyPending` is true and `autoVerify` is false**, and the clean-room path is available. **Recommended**, since verify is clean-room and rarely worth skipping. Runs verify now (require-clean mode); **does not** change config.
+- **Verify `{stage}` now + enable auto-verify going forward** — same trigger as the option above; runs verify now **and** patches `autoVerify: true` into `forge.config.json` in place (preserve formatting and other keys) so every future stage verifies automatically without prompting. This is the **folded** config-enable — it replaces the old post-hoc "make auto-verify the default?" follow-up. No silent config writes: the config changes **only** when the user picks this option.
+- **Verify `{stage}` first (manual)** — include **only when `verifyPending` is true** and the clean-room path is unavailable (the Tier-2 degradation above), or the host lacks the host's skill-invocation and subagent mechanisms; selecting it runs `{verifyCommand}` (inline / manual). Offer the "enable auto-verify" choice as text only if a config write is possible.
 - **Pick a different stage** — free-form escape to any stage or other action.
 
 **4. Act on the choice.**
-- **Start now** → if `autoInvokeNextStage` is true (default) **and** the `Skill` tool is available, invoke the chosen stage **via the `Skill` tool** in this same session (e.g. `skill: "feature-forge:forge-3-specs"`, `args: "{feature}"`) — no retyping, no paste. If `autoInvokeNextStage` is false, or the `Skill` tool is unavailable (a non-Claude host), fall back to printing `{nextCommand}` prominently for the user to run.
-- **Clean session** → give the exact next command and the `/clear`-then-re-run instruction; do not invoke anything.
-- **Verify** → invoke `feature-forge:forge-verify` via the `Skill` tool (or print `{verifyCommand}` on a non-Claude host).
+- **Clean session** (the default) → give the exact next command and the clear your session / start a fresh session-then-re-run instruction; do not invoke anything (you cannot clear your session / start a fresh session for the user).
+- **Continue in this session** → if `autoInvokeNextStage` is true (default) **and** the host's skill-invocation mechanism is available, invoke the chosen stage **via the host's skill-invocation mechanism** in this same session (e.g. `skill: "feature-forge:forge-3-specs"`, `args: "{feature}"`) — no retyping, no paste. If `autoInvokeNextStage` is false, or the host's skill-invocation mechanism is unavailable (a non-Claude host), fall back to printing `{nextCommand}` prominently for the user to run.
+- **Verify now / Verify now + enable auto-verify** → invoke `feature-forge:forge-verify` via the host's skill-invocation mechanism (require-clean mode). For the **enable-auto-verify** variant, additionally patch `autoVerify: true` into `forge.config.json` in place (preserve formatting and other keys) — never a silent write. On a non-Claude host or when degrading to the manual/inline gate, print `{verifyCommand}` instead.
 - **Different stage** → honor the free-form request.
 
+**Host fallback.** On a non-Claude host or when the host's skill-invocation and subagent mechanisms are unavailable, auto-verify never runs unattended — fall back to printing `{verifyCommand}` exactly as today, mirroring `autoInvokeNextStage`.
+
 This applies whether the feature was named explicitly (`/feature-forge:forge {feature}`) or resolved from the recency default.
 
 ### Epic Dashboard
@@ -118,7 +132,7 @@ This applies whether the feature was named explicitly (`/feature-forge:forge {fe
 When the named argument is an epic (`{specsDir}/{name}/epic-manifest.json` exists), render the epic dashboard instead of a per-feature one. Run:
 
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" render-status "{epic}" --specs-dir "{specsDir}" --json
 ```
@@ -180,16 +194,16 @@ Support these sub-commands for pipeline lifecycle management:
 - `/feature-forge:forge abandon {feature}` — Set `pipelineStatus` to `"abandoned"`. Use the host's question mechanism to confirm first, and state what's reversible: abandoning does not delete artifacts and can be undone with `/feature-forge:forge resume {feature}`, so the cost is low — but if the user really means "stop and discard," point out that `pause` is the better choice when they're only setting it aside. Offer **Abandon** · **Pause instead** · **Cancel**.
 - `/feature-forge:forge run [{feature}]` — **Opt-in auto-advance.** Drive the feature through consecutive stages in one session instead of confirming each boundary. This is a convenience wrapper over **3b. Drive to the Next Stage** — same stage order, same context gate — just looped:
   1. Resolve the feature (if omitted, use the recency default from `rank-features`; if multiple are equally plausible, ask once via the host's question mechanism).
-  2. **Before each stage,** run `forge-session.py context-usage`. If `overThreshold` is true, **stop** and recommend a clean session (give the exact `{nextCommand}` and the `/clear`-then-re-run instruction) — never auto-`/clear`.
-  3. Otherwise invoke the next stage's skill via the `Skill` tool, let it run to its natural stopping point, then re-read state and continue from step 2.
+  2. **Before each stage,** run `forge-session.py context-usage`. If `overThreshold` is true, **stop** and recommend a clean session (give the exact `{nextCommand}` and the clear your session / start a fresh session-then-re-run instruction) — never auto-clear your session / start a fresh session.
+  3. Otherwise invoke the next stage's skill via the host's skill-invocation mechanism, let it run to its natural stopping point, then re-read state and continue from step 2.
   4. **Stop conditions:** the next stage is an interview/decision point that calls the host's question mechanism (PRD and tech inherently pause for input — let them); `nextStage` is `null` (pipeline complete); context over threshold; or a stage signals it needs human input / is blocked. Always report where the loop stopped and why.
-  Per-stage confirmation (3b) remains the default — `run` is only used when the user explicitly asks to "run" / "drive" / "auto-advance" the pipeline. On a non-Claude host where the `Skill` tool is unavailable, fall back to printing the ordered list of commands to run.
+  Per-stage confirmation (3b) remains the default — `run` is only used when the user explicitly asks to "run" / "drive" / "auto-advance" the pipeline. On a non-Claude host where the host's skill-invocation mechanism is unavailable, fall back to printing the ordered list of commands to run.
 
 **Epic lifecycle.** When the argument names an **epic** (`{specsDir}/{name}/epic-manifest.json` exists), `pause` / `resume` / `abandon` operate on the epic manifest, not a `.pipeline-state.json`:
 
 - Set the manifest's top-level `status` (`paused` / `active` / `abandoned`) via the helper's `set-status` mutator — an atomic write that also bumps `updatedAt`:
   ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" set-status "{epic}" --status paused --specs-dir "{specsDir}"
   ```

package/adapters/codex/skills/forge-0-epic/SKILL.md

@@ -19,7 +19,7 @@ question in inline prose — every question goes through the host's question mec
 
 Read and follow `references/shared-conventions.md` for:
 - the **Feature Name Requirement** (applied here to the *epic* name — see below),
-- the **User Input Protocol** (the the host's question mechanism guardrail — all questions go through the tool),
+- the **User Input Protocol** (the host's question mechanism guardrail — all questions go through the tool),
 - **Configuration Reading**, and
 - the **Git Commit Protocol**.
 
@@ -38,7 +38,7 @@ checks but still load any on-disk artifacts.
 plugin path and the configured specs dir:
 
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" <subcommand> ... --specs-dir "{specsDir}"
 ```
@@ -72,14 +72,14 @@ Resolve the epic subtree path `{specsDir}/{epic}/` and decide which branch to ru
    epic, confirm the epic name itself does not collide with any existing feature or epic:
 
    ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" check-name "{epic}" --specs-dir "{specsDir}"
    ```
 
    - Exit `0` → the name is free; proceed to C1.
    - Exit `1` (`duplicate-name`) → STOP and surface the helper's finding **verbatim**; ask
-     (via the host's question mechanism) for a different epic name, then re-run check-name.
+     for a different epic name, then re-run check-name.
    - Exit `2` (unsafe name) → STOP and surface the finding; ask for a corrected name.
 
 ---
@@ -116,14 +116,14 @@ For **each** proposed feature name, before accepting it into the set, enforce gl
 and name safety via the helper:
 
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" check-name "{feature}" --specs-dir "{specsDir}"
 ```
 
 - Exit `0` → accept the name.
 - Exit `1` (`duplicate-name`) → reject that name; surface the finding verbatim and re-prompt
-  (via the host's question mechanism) for a different name.
+  for a different name.
 - Exit `2` (`unsafe-name`) → reject; surface the finding and re-prompt.
 
 Never accept a feature name that has not passed `check-name` exit 0.
@@ -176,7 +176,7 @@ For the *initial* creation write the skill writes the file directly — atomic g
 required for in-place mutation, which is the helper mutators' job. Creating the epic dir first creates `{specsDir}/`, so after writing the manifest invoke the **Specs Directory Hygiene** block in `references/shared-conventions.md` (idempotent; stage anything it writes with this stage's commit). Then validate:
 
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" validate "{epic}" --specs-dir "{specsDir}" --json
 ```
@@ -236,29 +236,24 @@ now self-contained: manifest + EPIC.md + one subdirectory per member.
 
 ### Step C8 — Review, Pipeline State & Commit
 
-1. **Review.** Present a summary (epic name, N features, dependency edges, contracts) as text,
-   then use the host's question mechanism: "Does this epic decomposition look right? Any feature, dependency,
-   or contract to change before I commit?" If the user wants changes, loop back to the relevant
-   creation step, re-compose, and re-validate.
+1. **Review.** Present a summary (epic name, N features, dependency edges, contracts) as text, then use the host's question mechanism: "Does this epic decomposition look right? Any feature, dependency, or contract to change before I commit?" If the user wants changes, loop back to the relevant creation step, re-compose, and re-validate.
 
-2. **Commit (Git Commit Protocol).** If `gitCommitAfterStage` is true, follow the Git Commit
-   Protocol in shared-conventions:
-   - Stage the whole epic subtree only: `git add {specsDir}/{epic}/` — never `git add -A`. This
-     captures `epic-manifest.json`, `EPIC.md`, and all member `.pipeline-state.json` files
-     atomically.
+2. **Commit (Git Commit Protocol).** If `gitCommitAfterStage` is true, follow the Git Commit Protocol in shared-conventions:
+   - Stage the whole epic subtree only: `git add {specsDir}/{epic}/` — never `git add -A`. This captures `epic-manifest.json`, `EPIC.md`, and all member `.pipeline-state.json` files atomically.
    - Commit with message `"{commitPrefix}({epic}): create epic with {N} features"`.
-   - On success, capture the commit hash. On failure (pre-commit hook, conflict), report and do
-     not mark complete; never use `--no-verify`/`--force`.
+   - On success, capture the commit hash for the closing message only — the epic manifest has no `commitHash` field, so nothing is written back into a committed file and the two-commit step of the Git Commit Protocol does not apply here. On failure (pre-commit hook, conflict), report and do not mark complete; never use `--amend`/`--no-verify`/`--force`.
 
-3. **Closing message.** After a successful creation, tell the user the next steps:
+3. **Closing message — the Stage Exit Protocol.** Congratulate the user ("Epic `{epic}` created with {N} features."), then close with the Stage Exit Protocol below (single-sourced in `references/stage-exit-protocol.md`; the epic → first-PRD boundary is a full stage boundary — do not improvise a "Next steps" list). `{first-actionable-feature}` = any feature with empty `dependsOn` (or the first entry of `render-status`'s `actionable` set):
 
-   > Epic `{epic}` created with {N} features. Next steps:
-   >  - `/feature-forge:forge {epic}` to see the epic dashboard
-   >  - `/feature-forge:forge-verify {epic}` to verify the epic
-   >  - `/feature-forge:forge-1-prd {first-actionable-feature}` to start the first feature
+**Close this stage with the Scripted Stage Exit** (contract: `references/stage-exit-protocol.md`; do not improvise a "Next steps" list). Run:
 
-   The first-actionable feature is any feature with empty `dependsOn` (or the first entry of
-   `render-status`'s `actionable` set).
+```bash
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
+[ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
+python3 "$R/scripts/forge-session.py" stage-exit --feature "{epic}" --stage forge-0-epic --next-feature "{first-actionable-feature}" --specs-dir "{specsDir}" --host generic
+```
+
+Obey the DIRECTIVES it prints, in order, per the directive contract: `runInStageVerify: true` → dispatch the in-stage clean-room verify now (honoring `autoFixEligible`); `verifyGate: "standard"` → present the Standard Verify Gate; `verifyGate: "manual-print"` → print the `verifyCommand` for the user; non-empty `invalidAutoVerifyKeys` → print a one-line warning. Then **print the NEXT-STEPS block verbatim as your absolute last output — nothing after its sentinel line.**
 
 ---
 

package/adapters/codex/skills/forge-0-epic/references/edit-mode.md

@@ -16,7 +16,7 @@ question goes through `AskUserQuestion`.
 Before offering any edit, validate the existing manifest:
 
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" validate "{epic}" --specs-dir "{specsDir}" --json
 ```
@@ -77,7 +77,7 @@ status is **not** `not-started`, warn the user. Read the **live** status (never
 completion in prose):
 
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" render-status "{epic}" --specs-dir "{specsDir}" --json
 ```
@@ -160,8 +160,10 @@ follow the Git Commit Protocol in shared-conventions:
    `"forge({epic}): create epic with 4 features"`, `"forge({epic}): add feature api-gateway"`,
    `"forge({epic}): remove feature legacy-session"`, `"forge({epic}): reorder features"`,
    `"forge({epic}): set dependency on token-service"`, or `"forge({epic}): set status paused"`.
-3. On success, capture the commit hash. On failure (pre-commit hook, conflict), report and do
-   **not** mark complete; never use `--no-verify`/`--force`.
+3. On success, capture the commit hash for reporting only — the epic manifest has no `commitHash`
+   field, so nothing is written back into a committed file and the two-commit step of the Git Commit
+   Protocol does not apply here. On failure (pre-commit hook, conflict), report and do **not** mark
+   complete; never use `--amend`/`--no-verify`/`--force`.
 
 Because every mutation is committed, the git history of `epic-manifest.json` is the audit trail; no
 separate in-manifest audit log is kept.

package/adapters/codex/skills/forge-0-epic/references/epic-manifest-subcommands.md

@@ -12,7 +12,7 @@ the write if it would introduce a cycle, dangling ref, duplicate, or schema viol
 flag surface (owned by 02 §7):
 
 ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 # Add a feature — seeds EMPTY exposes/consumes; contracts are populated below.
 python3 "$R/scripts/epic-manifest.py" add-feature "{epic}" "{feature}" \

package/adapters/codex/skills/forge-1-prd/SKILL.md

@@ -109,10 +109,19 @@ Write pipeline state conforming to `references/pipeline-state-schema.json`.
    - Record `artifacts`, `completedAt`
    - Set `stages.forge-1-prd.basedOnVersions` to `{}` (no upstream dependencies)
    - Check downstream stages (`forge-2-tech`, `forge-3-specs`, `forge-4-backlog`, `forge-5-loop`, `forge-6-docs`). If any have `basedOnVersions` referencing an older version of `forge-1-prd`, set their status to `stale`.
-2. Use the host's question mechanism to ask: "Anything you want to note before we wrap? (preserved across sessions)"
-   - If yes, store in the `notes` field
-3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol in `references/shared-conventions.md`: stage files (including `{specsDir}/AGENTS.md` / `{specsDir}/CLAUDE.md` if the Specs Directory Hygiene step just wrote them), attempt commit with message `"{commitPrefix}({feature}): complete PRD v{n}"`, then set `stages.forge-1-prd.status` to `complete` with commit hash only on success. If commit fails, leave status as `in-progress`.
-5. Tell the user: "PRD complete. Next steps:\n  - `/feature-forge:forge-verify {feature}` to verify the PRD\n  - `/feature-forge:forge-2-tech {feature}` to start the tech spec\n  - `/feature-forge:forge {feature}` to see full pipeline status"
+2. **Offer a note — don't force one.** As a statement (not a blocking question), let the user know they can jot anything worth preserving across sessions and you'll store it in the `notes` field. If they volunteer something, store it; otherwise proceed.
+3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol in `references/shared-conventions.md`: stage files (including `{specsDir}/AGENTS.md` / `{specsDir}/CLAUDE.md` if the Specs Directory Hygiene step just wrote them), attempt commit with message `"{commitPrefix}({feature}): complete PRD v{n}"` (marking `stages.forge-1-prd.status` `complete` with `commitHash: null` in that commit), then record the artifact-commit hash via the protocol's two-commit follow-up (never `--amend`) only on success. If commit fails, leave status as `in-progress`.
+4. **Close with the Stage Exit Protocol** (single-sourced in `references/stage-exit-protocol.md`; do not improvise a "Next steps" list):
+
+**Close this stage with the Scripted Stage Exit** (contract: `references/stage-exit-protocol.md`; do not improvise a "Next steps" list). Run:
+
+```bash
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
+[ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
+python3 "$R/scripts/forge-session.py" stage-exit --feature "{feature}" --stage forge-1-prd --specs-dir "{specsDir}" --host generic
+```
+
+Obey the DIRECTIVES it prints, in order, per the directive contract: `runInStageVerify: true` → dispatch the in-stage clean-room verify now (honoring `autoFixEligible`); `verifyGate: "standard"` → present the Standard Verify Gate; `verifyGate: "manual-print"` → print the `verifyCommand` for the user; non-empty `invalidAutoVerifyKeys` → print a one-line warning. Then **print the NEXT-STEPS block verbatim as your absolute last output — nothing after its sentinel line.**
 
 ## Gotchas
 

package/adapters/codex/skills/forge-2-tech/SKILL.md

@@ -87,7 +87,7 @@ Interview the user about technology decisions. Unlike the PRD interview, here yo
 - Recommend approaches consistent with the established stack, and say *why* the convention favors it (evidence-backed mode). Where the choice is genuine taste (e.g. folder layout, naming), give a default but flag it as preference.
 - Challenge over-engineering: does the feature need this, or is a simpler approach sufficient? Frame the simpler option's trade-off (less flexibility now vs. less to maintain).
 - Ask about every integration point and how the feature interacts with existing modules.
-- For competing module structures or code-shape choices, use the the host's question mechanism `preview` field to show the candidates side-by-side.
+- For competing module structures or code-shape choices, use the host's question mechanism `preview` field to show the candidates side-by-side.
 
 **Parking lot:** If the user raises a concern that belongs to a different pipeline stage (e.g., backlog granularity, documentation format), acknowledge it and note it in the pipeline state's `notes` field: "Good point — I've noted that for the [specs/backlog/docs stage]. Let's continue with the tech spec."
 
@@ -186,9 +186,19 @@ Write pipeline state conforming to `references/pipeline-state-schema.json`.
    - Record `artifacts`, `completedAt`, `version`
    - Set `stages.forge-2-tech.basedOnVersions` to `{"forge-1-prd": <current forge-1-prd version>}`
    - Check downstream stages (forge-3-specs, forge-4-backlog, forge-5-loop, forge-6-docs). If any have `basedOnVersions` referencing an older version of forge-2-tech, set their status to `stale`
-2. Use the host's question mechanism to ask about notes to persist
-3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol in `references/shared-conventions.md`: stage files, attempt commit with message `"{commitPrefix}({feature}): complete tech-spec v{n}"`, then set `stages.forge-2-tech.status` to `complete` with commit hash only on success. If commit fails, leave status as `in-progress`.
-5. Tell the user: "Tech spec complete. Next steps:\n  - `/feature-forge:forge-verify {feature}` to verify the tech spec\n  - `/feature-forge:forge-3-specs {feature}` to create implementation specs\n  - `/feature-forge:forge {feature}` to see full pipeline status"
+2. **Offer a note — don't force one.** As a statement (not a blocking question), let the user know they can jot anything worth preserving across sessions and you'll store it in the `notes` field. If they volunteer something, store it; otherwise proceed.
+3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol in `references/shared-conventions.md`: stage files, attempt commit with message `"{commitPrefix}({feature}): complete tech-spec v{n}"` (marking `stages.forge-2-tech.status` `complete` with `commitHash: null` in that commit), then record the artifact-commit hash via the protocol's two-commit follow-up (never `--amend`) only on success. If commit fails, leave status as `in-progress`.
+4. **Close with the Stage Exit Protocol** (single-sourced in `references/stage-exit-protocol.md`; do not improvise a "Next steps" list):
+
+**Close this stage with the Scripted Stage Exit** (contract: `references/stage-exit-protocol.md`; do not improvise a "Next steps" list). Run:
+
+```bash
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
+[ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
+python3 "$R/scripts/forge-session.py" stage-exit --feature "{feature}" --stage forge-2-tech --specs-dir "{specsDir}" --host generic
+```
+
+Obey the DIRECTIVES it prints, in order, per the directive contract: `runInStageVerify: true` → dispatch the in-stage clean-room verify now (honoring `autoFixEligible`); `verifyGate: "standard"` → present the Standard Verify Gate; `verifyGate: "manual-print"` → print the `verifyCommand` for the user; non-empty `invalidAutoVerifyKeys` → print a one-line warning. Then **print the NEXT-STEPS block verbatim as your absolute last output — nothing after its sentinel line.**
 
 ## Gotchas
 

package/adapters/codex/skills/forge-3-specs/SKILL.md

@@ -140,9 +140,19 @@ Write pipeline state conforming to `references/pipeline-state-schema.json`.
    - Record all created files in `artifacts`, including `TRACEABILITY.md`
    - Set `stages.forge-3-specs.basedOnVersions` to `{"forge-1-prd": <current version>, "forge-2-tech": <current version>}`
    - Check downstream stages (forge-4-backlog, forge-5-loop, forge-6-docs). If any have `basedOnVersions` referencing older versions, set their status to `stale`
-2. Use the host's question mechanism to ask about notes to persist
-3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol in `references/shared-conventions.md`: stage files, attempt commit with message `"{commitPrefix}({feature}): complete implementation specs v{n}"`, then set `stages.forge-3-specs.status` to `complete` with commit hash only on success. If commit fails, leave status as `in-progress`.
-5. Tell the user: "Implementation specs complete. Next steps:\n  - `/feature-forge:forge-verify {feature}` to verify the specs (strongly recommended)\n  - `/feature-forge:forge-4-backlog {feature}` to generate the backlog\n  - `/feature-forge:forge {feature}` to see full pipeline status"
+2. **Offer a note — don't force one.** As a statement (not a blocking question), let the user know they can jot anything worth preserving across sessions and you'll store it in the `notes` field. If they volunteer something, store it; otherwise proceed.
+3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol in `references/shared-conventions.md`: stage files, attempt commit with message `"{commitPrefix}({feature}): complete implementation specs v{n}"` (marking `stages.forge-3-specs.status` `complete` with `commitHash: null` in that commit), then record the artifact-commit hash via the protocol's two-commit follow-up (never `--amend`) only on success. If commit fails, leave status as `in-progress`.
+4. **Close with the Stage Exit Protocol** (single-sourced in `references/stage-exit-protocol.md`; do not improvise a "Next steps" list). Specs feed every downstream stage, so the verify gate matters here:
+
+**Close this stage with the Scripted Stage Exit** (contract: `references/stage-exit-protocol.md`; do not improvise a "Next steps" list). Run:
+
+```bash
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
+[ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
+python3 "$R/scripts/forge-session.py" stage-exit --feature "{feature}" --stage forge-3-specs --specs-dir "{specsDir}" --host generic
+```
+
+Obey the DIRECTIVES it prints, in order, per the directive contract: `runInStageVerify: true` → dispatch the in-stage clean-room verify now (honoring `autoFixEligible`); `verifyGate: "standard"` → present the Standard Verify Gate; `verifyGate: "manual-print"` → print the `verifyCommand` for the user; non-empty `invalidAutoVerifyKeys` → print a one-line warning. Then **print the NEXT-STEPS block verbatim as your absolute last output — nothing after its sentinel line.**
 
 ## Gotchas
 

package/adapters/codex/skills/forge-4-backlog/SKILL.md

@@ -38,7 +38,7 @@ Resolve the **loop runner** from the `loopRunner` block in `forge.config.json`,
 
 **Prerequisite check:** Read `{resolvedFeatureDir}/.pipeline-state.json`. If not in force mode, stages `forge-1-prd`, `forge-2-tech`, and `forge-3-specs` must all be `complete`. If not, STOP and tell the user which prerequisites are missing.
 
-**Strongly recommended:** Check if specs have been verified. If not, use the host's question mechanism to warn with the cost of skipping: "Specs haven't been verified yet. Recommended: run `/feature-forge:forge-verify {feature}` first — unverified specs can carry gaps or contradictions that get baked into backlog items and only surface mid-loop, where they're far more expensive to fix. Continue anyway?" Offer **Verify first (recommended)** · **Continue without verifying**.
+**Verification check.** Check whether the specs have been verified. If not, use the host's question mechanism to warn with the cost of skipping: "Specs haven't been verified yet. Recommended: run `/feature-forge:forge-verify {feature}` first — unverified specs can carry gaps or contradictions that get baked into backlog items and only surface mid-loop, where they're far more expensive to fix. Continue anyway?" Offer **Verify first (recommended)** · **Continue without verifying**.
 
 ## Step 2: Load All Specs
 
@@ -122,7 +122,7 @@ Interpret the result:
 
 Present a summary: total items N, dependency-chain depth, estimated loop iterations (`ceil(pendingItems * loopIterationMultiplier)`). Note whether validation passed or was skipped (runner not yet available).
 
-Use the host's question mechanism to ask: "Ready to proceed, or any adjustments needed?"
+State that the backlog is ready and invite adjustments before committing — a statement, not a forced gate: "Backlog is ready. Tell me if you want any items split, merged, or reordered; otherwise I'll record state and commit." Proceed to Step 7 unless the user asks for changes.
 
 ## Step 7: Update Pipeline State and Commit
 
@@ -133,10 +133,20 @@ Write pipeline state conforming to `references/pipeline-state-schema.json`. Foll
    - Set `stages.forge-4-backlog.basedOnVersions` to `{"forge-1-prd": <current version>, "forge-2-tech": <current version>, "forge-3-specs": <current version>}`
    - Set `currentStage` to `forge-5-loop`
    - Check downstream stages (`forge-5-loop`, `forge-6-docs`). If any have `basedOnVersions` referencing an older version of `forge-4-backlog`, set their status to `stale`.
-2. Use the host's question mechanism to ask about notes to persist
-3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol: stage files, attempt commit, then set status to `complete` with commit hash only on success. If commit fails, leave status as `in-progress`.
+2. **Offer a note — don't force one.** As a statement (not a blocking question), let the user know they can jot anything worth preserving across sessions and you'll store it in the `notes` field. If they volunteer something, store it; otherwise proceed.
+3. If `gitCommitAfterStage` is true, follow the Git Commit Protocol: stage files, attempt commit (marking `stages.forge-4-backlog.status` `complete` with `commitHash: null` in that commit), then record the artifact-commit hash via the protocol's two-commit follow-up (never `--amend`) only on success. If commit fails, leave status as `in-progress`.
 4. If verification was available but the user chose to skip it, record `stages.forge-verify-backlog.status` as `"skipped"` in pipeline state.
-5. Tell user: "Backlog complete with {N} items. Next steps:\n  - `/feature-forge:forge-verify {feature}` to verify the backlog\n  - `/feature-forge:forge-5-loop {feature}` to run the loop\n  - `/feature-forge:forge {feature}` to see full pipeline status"
+5. **Close with the Stage Exit Protocol** (single-sourced in `references/stage-exit-protocol.md`; do not improvise a "Next steps" list). Lead with the item count ("Backlog complete with {N} items."), then:
+
+**Close this stage with the Scripted Stage Exit** (contract: `references/stage-exit-protocol.md`; do not improvise a "Next steps" list). Run:
+
+```bash
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
+[ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
+python3 "$R/scripts/forge-session.py" stage-exit --feature "{feature}" --stage forge-4-backlog --specs-dir "{specsDir}" --host generic
+```
+
+Obey the DIRECTIVES it prints, in order, per the directive contract: `runInStageVerify: true` → dispatch the in-stage clean-room verify now (honoring `autoFixEligible`); `verifyGate: "standard"` → present the Standard Verify Gate; `verifyGate: "manual-print"` → print the `verifyCommand` for the user; non-empty `invalidAutoVerifyKeys` → print a one-line warning. Then **print the NEXT-STEPS block verbatim as your absolute last output — nothing after its sentinel line.**
 
 ## Gotchas
 

package/adapters/codex/skills/forge-5-loop/SKILL.md

@@ -49,7 +49,7 @@ Read `{resolvedFeatureDir}/.pipeline-state.json`. If not in force mode, `stages.
 
 ### 1b. Verification Check
 
-Check if `stages.forge-verify-backlog` exists and has status `passed` or `findings-applied`. If not, use the host's question mechanism to warn:
+Check if `stages.forge-verify-backlog` exists and has status `passed` or `findings-applied`. If not, use the host's question mechanism to warn with the cost of skipping:
 
 "Backlog hasn't been verified yet. Recommended: run `/feature-forge:forge-verify {feature}` first — the loop implements items autonomously and commits as it goes, so a bad item (wrong scope, missing dependency, untestable acceptance criteria) is far cheaper to catch now than after several commits build on it. Continue anyway?" Offer **Verify first (recommended)** · **Continue without verifying**.
 
@@ -60,7 +60,7 @@ Read the resolved feature's `.pipeline-state.json`. **If it has no `epic` key, s
 1. Run `render-status "{epic}" --specs-dir "{specsDir}" --json` via the helper:
 
    ```bash
-R="$(for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done)"
+R="$(bash -c 'for d in "$HOME"/.claude/skills/feature-forge "$HOME"/.claude/plugins/cache/*/feature-forge/* "$HOME"/.claude/plugins/*/feature-forge "$HOME"/.agents/skills/feature-forge ./.agents/skills/feature-forge; do [ -x "$d/scripts/forge-root.sh" ] && exec "$d/scripts/forge-root.sh"; done')"
 [ -n "$R" ] || { echo "feature-forge: cannot locate plugin root" >&2; exit 1; }
 python3 "$R/scripts/epic-manifest.py" \
   render-status "{epic}" --specs-dir "{specsDir}" --json
@@ -210,7 +210,7 @@ verbatim "Loop started…" inform-user output template is in
 
 ### 3d. Arm a Monitor on the event stream, and react to events
 
-Arm the **the host's monitoring mechanism** on the structured event stream (the NDJSON file, or the
+Arm the **host's monitoring mechanism** on the structured event stream (the NDJSON file, or the
 human log as fallback) so events flow back into this session as they happen. Use
 **`persistent: true`** — runs can exceed the host's monitoring mechanism's maximum `timeout_ms` (1 hour),
 and a bounded timeout would silently stop watching a still-running loop. The filter
@@ -258,11 +258,7 @@ output templates — **all-done**, **needs-human**, **blocked**, **deferred**, a
 
 Update `{resolvedFeatureDir}/.pipeline-state.json`:
 
-1. Set `stages.forge-5-loop`:
-   - `status`: `"complete"` if all backlog items are `done`, otherwise `"in-progress"`
-   - `completedAt`: current ISO timestamp (only if complete)
-   - `basedOnVersions`: `{"forge-4-backlog": <current version from pipeline state>}`
-   - `artifacts`: `["{backlogDir}/{loopRunner.stateDir}/state.json"]`
+1. Set `stages.forge-5-loop`: `status` = `"complete"` if all backlog items are `done`, else `"in-progress"`; `completedAt` = current ISO timestamp (only if complete); `basedOnVersions` = `{"forge-4-backlog": <current version from pipeline state>}`; `artifacts` = `["{backlogDir}/{loopRunner.stateDir}/state.json"]`.
 2. If all items complete: set `currentStage` to `"forge-6-docs"`
 3. Update `updatedAt`
 
@@ -276,30 +272,32 @@ Update `{resolvedFeatureDir}/.pipeline-state.json`:
 
 **Gate:** only run this step if (a) the resolved feature's `.pipeline-state.json` has an `epic` key **and** (b) Step 5 set `stages.forge-5-loop.status` to `complete` (all backlog items done). If either is false, **skip** — standalone completed features are handled by Step 5b, and partial runs end as today (REQ-COMPAT-01).
 
-1. **Offer impl-verify first (recommended, skippable).** Per the completion rule (`00-core-definitions.md §7`), a feature whose `forge-verify-impl.status == findings-reported` does **not** unblock dependents. Use the host's question mechanism (NOT inline prose) to offer:
-
-   > "{feature}'s loop is done. Recommended: run `/feature-forge:forge-verify {feature} impl` before unblocking dependents. Run it now, or skip and continue the handoff?"
-
-   The user may skip (then completion is judged on the §7 rule with impl-verify absent).
+1. **Offer impl-verify first (recommended, skippable).** Per the completion rule (`00-core-definitions.md §7`), a feature whose `forge-verify-impl.status == findings-reported` does **not** unblock dependents. Use the host's question mechanism (NOT inline prose) to offer: *"{feature}'s loop is done. Recommended: run `/feature-forge:forge-verify {feature} impl` before unblocking dependents. Run it now, or skip and continue the handoff?"* The user may skip (then completion is judged on the §7 rule with impl-verify absent).
 2. **Recompute and announce.** Run `render-status "{epic}" --specs-dir "{specsDir}" --json`. Announce the feature's completion and the epic rollup (e.g. "2/4 features complete") — derived live from disk, never re-computed in prose.
-3. **Identify the next actionable feature(s).** Read `render-status`'s `actionable` set (features whose every dependency is now complete and that are not themselves complete) and `nextCommand`.
-   - **None actionable** (everything done, or remaining features still blocked): say so.
-     - If `rollup.total > 0` **AND** `rollup.complete == rollup.total`, suggest `/feature-forge:forge-6-docs {feature}` and note the epic-level documentation offer (§10). The `rollup.total > 0` guard prevents an **empty epic** (`0 == 0`) from being reported complete.
-     - Otherwise, list what is still blocked and on which dependencies. End — do not prompt to start a feature that cannot start.
-   - **One or more actionable:** use the host's question mechanism presenting **each actionable feature** as an option (plus "stop here"). Execution is **serial** — the user picks exactly one (REQ-ORCH-03). Do **not** autonomously chain into the next pipeline.
-4. **Begin the chosen feature.** For the picked feature:
-   - **PRD absent** (no `PRD.md`, or `stages.forge-1-prd` not complete): offer to author it now — "Start `/feature-forge:forge-1-prd {chosen}`?" (REQ-ORCH-02). On yes, hand off to forge-1-prd (which injects epic context per §5.1).
-   - **PRD present:** point the user at the chosen feature's `nextCommand` from render-status.
+3. **Identify the next actionable feature(s).** Read `render-status`'s `actionable` set (every dependency now complete, not itself complete) and `nextCommand`. **None actionable:** say so — if `rollup.total > 0` **AND** `rollup.complete == rollup.total`, suggest `/feature-forge:forge-6-docs {feature}` and note the epic-level documentation offer (§10) (the `rollup.total > 0` guard prevents an **empty epic** `0 == 0` from reading as complete); otherwise list what is still blocked and on which dependencies, then end (do not prompt to start a feature that cannot start). **One or more actionable:** use the host's question mechanism presenting **each actionable feature** as an option (plus "stop here"). Execution is **serial** — the user picks exactly one (REQ-ORCH-03); do **not** autonomously chain into the next pipeline.
+4. **Begin the chosen feature.** **PRD absent** (no `PRD.md`, or `stages.forge-1-prd` not complete): offer to author it now — "Start `/feature-forge:forge-1-prd {chosen}`?" (REQ-ORCH-02); on yes, hand off to forge-1-prd (which injects epic context per §5.1). **PRD present:** point the user at the chosen feature's `nextCommand` from render-status.
 5. **Commit (REQ-OBS-01).** When `gitCommitAfterStage` is true, commit the Step 5 completion write (and any manifest `updatedAt` bump) via the shared-conventions **Git Commit Protocol**, staging the epic subtree so the member state change commits atomically: `git add {specsDir}/{epic}/` then `{commitPrefix}({feature}): complete loop`. If `gitCommitAfterStage` is false, skip the commit.
+6. **Close the handoff with the Stage Exit Protocol.** Finishing feature `{feature}` → starting the picked feature `{chosen}`'s PRD is a **cold** stage boundary (single-sourced in `references/stage-exit-protocol.md`). Feature `{feature}`'s impl-verify was already offered in Step 6.1, so step 1's gate self-suppresses when it ran or was skipped — the block collapses to clear your session / start a fresh session → next-command. Present it only when the chosen feature's PRD is absent (a PRD-present pick just runs its `nextCommand`):
+
+**This stage is done — walk the user through the Stage Exit Protocol** before moving on. The order is fixed, and step 2 is something only the user can do:
+
+1. **Verify feature {feature}'s loop first — if it isn't already verified.** If verify already ran in this session — via the in-stage auto-verify on the authoring stages, or the interactive impl-verify offered above on the loop — or is already fresh on record, or the stage was explicitly skipped, say so and go straight to step 2. Only when `autoVerify` is off for this stage **and** verify is **missing or stale** do you present the **Standard Verify Gate**: verify **now, before clearing**, using the host's question mechanism with exactly these three options — but only when the host has a question mechanism **and** the clean-room path is available (the host's subagent mechanism plus a dispatchable `forge-verifier` subagent):
+   - **Verify feature {feature}'s loop now** *(recommended)* — dispatch the clean-room `forge-verifier` subagent from this session in require-clean mode; the digest returns here so any fix decision keeps its context. One-time — it does **not** change config.
+   - **Verify now + enable auto-verify going forward** — verify now **and** patch `"autoVerify": true` into `forge.config.json` in place (preserve formatting and every other key) so future stages verify automatically, no prompt. This complements the `forge-init` opt-in. **Do not auto-commit this config change** — treat it like `notes`: a user-facing edit the user commits on their own cadence, never folded into a stage's artifact commit.
+   - **Skip for now** — go straight to clear your session / start a fresh session and the next command without verifying. Record this stage's verify status as `"skipped"` in pipeline state (mirroring the existing skip handling) **only** on an explicit skip — a skip does not go stale.
+
+   **Host / clean-room fallback (not a user-selectable option):** if the question mechanism, the host's subagent mechanism, or the `forge-verifier` subagent is unavailable, do **not** run clean-room — degrade to printing `/feature-forge:forge-verify {feature} impl` for the user to run inline/manually (mirroring `autoInvokeNextStage`), and offer the auto-verify enable as plain text only if a config write is possible.
+2. **Then clear your session / start a fresh session.** Recommended **unconditionally** at this boundary for a clean start — independent of how full the context window is. Every artifact is on disk, so the work survives the clear. **I can't clear your session / start a fresh session for you — you have to run it yourself.**
+3. **Then run `/feature-forge:forge-1-prd {chosen}`** in the fresh session — or re-run `/feature-forge:forge` to let the navigator resume from disk.
 
 ## Gotchas
 
-- **Plugin-root discovery (1b-epic helper) covers installed paths, not workspace-dev checkouts.** The `forge-root.sh` search in 1b-epic probes `~/.claude/skills/feature-forge`, `~/.claude/plugins/*/feature-forge`, and `./.agents/skills/feature-forge` — the locations of an **installed** plugin. A feature-forge **source checkout** (e.g. `~/workspace/feature-forge`) is not on that list, so the helper exits "cannot locate plugin root." That is expected in a dev environment, not a bug; run the epic-manifest script from the checkout directly (`python3 <checkout>/scripts/epic-manifest.py …`). Note the `~/.claude/plugins/*/feature-forge` glob can also print a zsh "no matches found" line when that dir is empty — harmless.
+- **Plugin-root discovery (1b-epic helper) covers installed paths, not workspace-dev checkouts.** The `forge-root.sh` search in 1b-epic probes `~/.claude/skills/feature-forge`, `~/.claude/plugins/cache/*/feature-forge/*` (marketplace-cache installs), `~/.claude/plugins/*/feature-forge`, and `./.agents/skills/feature-forge` — the locations of an **installed** plugin. A feature-forge **source checkout** (e.g. `~/workspace/feature-forge`) is not on that list, so the helper exits "cannot locate plugin root." That is expected in a dev environment, not a bug; run the epic-manifest script from the checkout directly (`python3 <checkout>/scripts/epic-manifest.py …`). The bootstrap prelude wraps its candidate loop in `bash -c` so the `~/.claude/plugins/*/feature-forge` glob is zsh-safe: an empty expansion no longer aborts the loop under zsh's `nomatch`.
 - `{backlogDir}` is a **directory path**, not a file path. Pass `specs/auth`, not `specs/auth/backlog.json`.
 - rauf resolves `RAUF.md` with fallback: checks `{backlogDir}/.rauf/RAUF.md` first, then the project's `.rauf/RAUF.md`. As long as the runner is installed in the project, the prompt template will be found.
 - State files (state.json, {loopRunner.logFile}, etc.) are created at `{backlogDir}/{loopRunner.stateDir}/` — this is within the feature's spec directory and is expected. State is isolated per backlog dir, so concurrent features don't collide.
 - If the session disconnects during a long-running loop, the runner process continues independently. The user can check results later with the status / list commands.
-- Never run the run command in the foreground (without the host's background-execution mechanism) — it blocks and will hit the Bash tool timeout for any non-trivial backlog. "Don't block the foreground" is NOT "stay silent": supervise via the host's monitoring mechanism (3d), never `sleep`/poll in the foreground. The the host's monitoring mechanism must use `persistent: true` (not a bounded `timeout_ms`), watch the **structured** surface (`events.ndjson`), and never filter on raw `RAUF_*` tokens — they appear in agent prose and false-match. A `needs_human`/`blocked`/`review` signal does **not** pause the loop — the runner sets the item aside and keeps going; surface it live but don't tell the user the loop is waiting. See `references/runner-contract.md` for the full monitoring rules.
+- Never run the run command in the foreground (without the host's background-execution mechanism) — it blocks and will hit the Bash tool timeout for any non-trivial backlog. "Don't block the foreground" is NOT "stay silent": supervise via the host's monitoring mechanism (3d), never `sleep`/poll in the foreground. The host's monitoring mechanism must use `persistent: true` (not a bounded `timeout_ms`), watch the **structured** surface (`events.ndjson`), and never filter on raw `RAUF_*` tokens — they appear in agent prose and false-match. A `needs_human`/`blocked`/`review` signal does **not** pause the loop — the runner sets the item aside and keeps going; surface it live but don't tell the user the loop is waiting. See `references/runner-contract.md` for the full monitoring rules.
 - If a previous loop run left a stale lock, the user may need to pass `--force` to clear it. rauf will report this error clearly.
 - The version gate (1c) uses the `--json` form on purpose; never parse `rauf version`'s human output.
 - **Implementation artifacts must not cite specs.** The loop should **read** the specs and `backlog.json` freely — they are the source of truth for what to build, and the backlog rightly references specs for provenance. But the artifacts the loop **writes into the target repo** (source code, generated `SKILL.md`/agent files, configs, code comments) must be **self-contained**: they must NOT reference feature-forge spec files (no `See specs/{feature}/NN-*.md`, no "source spec" provenance notes in shipped output). Specs are pre-implementation inputs that may be archived or deleted once the feature ships; the implementation must stand on its own. This applies only to shipped implementation output — never to the backlog or spec documents, which should keep citing specs.