@open-agent-toolkit/cli 0.2.56 → 0.2.57

package/assets/agents/oat-phase-implementer.md

@@ -1,6 +1,6 @@
 ---
 name: oat-phase-implementer
-version: 1.1.1
+version: 1.1.2
 description: Implements one plan phase end-to-end, commits each task separately, self-checks between tasks, and handles bounded review fixes when resumed by oat-project-implement.
 tools: Read, Write, Edit, Bash, Grep, Glob, Task
 color: cyan
@@ -37,7 +37,8 @@ The root supplies:
   candidates, and formal dispatch stamp;
 - `phase_recovery_limit`, `phase_recovery_attempts_used`,
   `original_request_id`, and the root-resolved recovery authorization source;
-- optional `parallel_group`, `expected_base_sha`, and smoke run metadata.
+- optional `parallel_group`, `expected_base_sha`, `recovered_patch`, and
+  smoke run metadata.
 
 Fix mode also supplies:
 
@@ -341,6 +342,16 @@ use ancestry from `expected_base_sha` as a substitute for the exact
 `phase_base_head` check. For a plan-declared parallel group, verify this before
 any task edit.
 
+A fresh child never starts on a dirty tree. The one exception is a supplied
+`recovered_patch`, which carries work a lost child left behind: verify it,
+apply it, review it, and commit it as your first action before task one,
+exactly as `## Mode: Recover` steps 2 and 4 describe, and only after every
+precondition that can park the phase has passed. That single commit is a
+recovery commit rather than a planned task commit, so it does not consume any
+task's one-commit budget; report it under `### Recovery Events` with the
+artifact reference and manifest digest. Without a verified `recovered_patch`,
+any dirt blocks.
+
 When smoke containment, ownership registration, expected base, or fixture
 readiness proves the run invalid, return `INVALID_RUN_ABORT` with the evidence.
 Do not launch a child, continue sequentially, review, or repair the invalid run.
@@ -462,8 +473,18 @@ phase_verification: { relevant phase command }
 dispatch_target: { exact original launcher-owned target }
 dispatch_axes: { unchanged original launcher-owned axes }
 dispatch_stamp: { original formal Dispatch line }
+recovered_patch: { artifact, manifest_digest, size, stat, components }
 ```
 
+`recovered_patch` is optional. It is present only when a lost child left
+uncommitted work that the lifecycle captured before restoring the worktree to
+its clean base. Its `artifact` is a readable path outside the worktree, never a
+mutable worktree path; `manifest_digest` and `size` are the capture's recorded
+values; `stat` is the captured `git diff --cached --stat` summary; and
+`components` lists the `index`, `worktree`, and `untracked` entries the
+artifact holds. The same field may arrive on a same-target `implement`
+continuation.
+
 1. Validate every recover input. Confirm the exact launcher-owned target equals
    the original target and the generic `continuation_events` record links
    `continuation_event` to `original_request_id`.
@@ -471,23 +492,116 @@ dispatch_stamp: { original formal Dispatch line }
    remains immutable at the same history position; and the worktree contains
    only the reconciled pending ledger reservation plus an optional mechanically
    bounded diff inside `bounded_files`. Any other dirt or history change
-   blocks.
+   blocks. Compare against the briefed `recovery_base_head`, never a remembered
+   earlier value: once a recovery-patch commit has landed, the retry's
+   `recovery_base_head` is the reconciled current HEAD, which includes that
+   recovery commit and any candidate or ledger-only commit already durably
+   made, and the brief omits the already-committed artifact because re-applying
+   it would duplicate the work. Beyond that reconciled reservation and its
+   bounded diff, a supplied `recovered_patch` is the only permitted
+   pre-existing dirt, and only after it verifies. Verification is read-only and
+   happens here; the apply and the commit wait for step 4, so every precondition
+   that can park the phase resolves before anything is written. When the brief
+   carries no `recovered_patch` — an ordinary resume, or a retry after the
+   recovery commit already landed — the ordinary permitted-dirt invariant above
+   is the whole of this step: there is no artifact to resolve a script for, so
+   the continuation skips verification and continues to the ledger
+   reconciliation in step 3, and step 4 is a no-op. Resolve
+   `capture-dirty-tree.mjs` through installed scope rather than a
+   repository-relative path, probing in order and binding the first match, and
+   only when an artifact was briefed:
+
+   ```bash
+   set -eu
+   # `recovered_patch` is optional and all-or-nothing. Without one there is
+   # nothing to verify and nothing to apply, so the verifier is skipped
+   # entirely; a partial brief is a named stop rather than an unbound-variable
+   # death. Never dereference an artifact variable outside the branch: under
+   # `set -u` an unset one exits here, and an empty one fails verification —
+   # either way an artifact-free retry parks before the ledger reconciliation
+   # it needs to reach.
+   if [ -z "${ARTIFACT_DIR:-}${MANIFEST_DIGEST:-}${ARTIFACT_SIZE:-}" ]; then
+     : # No recovered_patch: continue to the ledger reconciliation.
+   elif [ -z "${ARTIFACT_DIR:-}" ] || [ -z "${MANIFEST_DIGEST:-}" ] ||
+     [ -z "${ARTIFACT_SIZE:-}" ]; then
+     echo "artifact-verification-failed: a partial recovered_patch is unusable; artifact, manifest_digest and size travel together" >&2
+     exit 1
+   else
+     CAPTURE_SCRIPT=""
+     REPO_ROOT=$(git -C "$PHASE_WORKTREE" rev-parse --show-toplevel 2>/dev/null || true)
+     for CAPTURE_ROOT in "${HOME:-}/.agents/skills/oat-project-implement" \
+       "${REPO_ROOT:+$REPO_ROOT/.agents/skills/oat-project-implement}"; do
+       [ -n "$CAPTURE_ROOT" ] || continue
+       [ -f "$CAPTURE_ROOT/scripts/capture-dirty-tree.mjs" ] || continue
+       CAPTURE_SCRIPT="$CAPTURE_ROOT/scripts/capture-dirty-tree.mjs"
+       break
+     done
+     [ -n "$CAPTURE_SCRIPT" ] || {
+       echo "capture-script-unavailable" >&2
+       exit 1
+     }
+
+     EXPECTED_HEAD=$(git -C "$PHASE_WORKTREE" rev-parse --verify HEAD)
+     node "$CAPTURE_SCRIPT" --verify \
+       --artifact-dir "$ARTIFACT_DIR" \
+       --manifest-digest "$MANIFEST_DIGEST" \
+       --size "$ARTIFACT_SIZE" \
+       --expected-head "$EXPECTED_HEAD"
+   fi
+   ```
+
+   `ARTIFACT_DIR`, `MANIFEST_DIGEST`, and `ARTIFACT_SIZE` are the
+   `recovered_patch` fields; bind them as shell variables rather than pasting
+   `<placeholder>` tokens, which a shell reads as input redirections. The guard
+   terminates rather than warning: `node ""` reads its program from stdin and
+   exits zero at EOF, so a fall-through would report an unverified artifact as
+   verified. Every block that runs the script repeats the resolution and the
+   guard, because shell variables do not survive across separate tool
+   invocations. `--expected-head` and the
+   apply below both resolve against `$PHASE_WORKTREE`, never the ambient
+   working directory. An empty `CAPTURE_SCRIPT` is a
+   `capture-script-unavailable` stop: report it verbatim with the probed roots
+   and the recovery command
+   `oat tools install workflows --scope <user|project>` or
+   `oat tools update --pack workflows --scope <user|project>`, and never fall
+   back to a repository-relative path. With an artifact briefed, verify before
+   touching the tree and stop on any `artifact-verification-failed` mismatch, including the base mismatch
+   `--expected-head` catches: integrity is not base agreement, and a hunk whose
+   context happens to match applies cleanly at the wrong commit. Report an
+   `active-writer`, `unsupported-dirt`, `round-trip-failed`, or
+   `artifact-verification-failed` reason verbatim and stop; never improvise a
+   partial or best-effort restore. Without a verified `recovered_patch` you
+   start on a clean tree.
+
 3. Reconcile the authoritative `pending_attempt` and nonzero
    `phase_recovery_attempts_used` with `state.md`. Recover mode continues that
    same consumed attempt and must not increment usage again. Missing,
-   contradictory, or unreconciled state blocks before editing.
-4. Apply or complete only `bounded_correction_scope`. Recover mode must not
+   contradictory, or unreconciled state blocks before editing. Nothing has been
+   applied or committed yet, so this block leaves the branch exactly at
+   `recovery_base_head` with the artifact untouched and re-verifiable; the root
+   re-briefs from current authoritative state once the ledger is repaired.
+4. Only now apply the verified `recovered_patch`, if one was supplied:
+   `git apply --index` for the `index` component, `git apply` for the
+   `worktree` component, and byte copies plus the manifest's recorded
+   executable bit for the `untracked` component. Review the result and commit
+   it as your first action. That commit is immutable, and so is every commit
+   this mode durably makes after it: if anything later parks the phase, the
+   retry's `recovery_base_head` is the reconciled current HEAD, which includes
+   that recovery commit and any candidate or ledger-only commit already made.
+   That brief omits the already-committed artifact; dirt that accumulated
+   afterwards travels as its own newly captured `recovered_patch`.
+5. Apply or complete only `bounded_correction_scope`. Recover mode must not
    replay planned tasks and must not require, fabricate, or consume a review
    artifact.
-5. Apply the bounded correction, then run `focused_verification` and
+6. Apply the bounded correction, then run `focused_verification` and
    `phase_verification` before creating a candidate commit. If either check
    fails, restore `bounded_files` to their original committed content,
    atomically mark the pending entry `failed`, durably commit only that
    ledger transition, emit one `failed-attempt` event, and stop.
-6. When both pre-commit checks pass, atomically mark the pending entry
+7. When both pre-commit checks pass, atomically mark the pending entry
    `completed` and create one append-only candidate recovery commit containing
    only `bounded_files` plus that ledger transition.
-7. Immediately rerun `focused_verification` and `phase_verification` against the
+8. Immediately rerun `focused_verification` and `phase_verification` against the
    committed HEAD; these reruns are authoritative. On pass, emit one `recovered`
    event and report the candidate as the successful recovery commit. On
    failure, atomically replace `completed` with `failed`, durably commit that
@@ -496,7 +610,7 @@ dispatch_stamp: { original formal Dispatch line }
    commit. If terminal evidence cannot be committed, report an unreconciled
    block that root must reject before bookkeeping. Never amend history or
    launch fallback.
-8. Return the report below. `DONE` is accepted success;
+9. Return the report below. `DONE` is accepted success;
    `DONE_WITH_CONCERNS` and `BLOCKED` may be accepted terminal stops and must
    still report provenance, accounting, immutable history, and the event.
 

package/assets/bundle-metadata.json

@@ -1,4 +1,4 @@
 {
   "schemaVersion": 1,
-  "oatVersion": "0.2.56"
+  "oatVersion": "0.2.57"
 }

package/assets/docs/workflows/projects/implementation-execution.md

@@ -152,7 +152,11 @@ by the phase's standing authority. The important distinctions are:
   boundary, not for every eligible defect;
 - a **continuation** preserves the original request and exact target, either in
   the accepted handle or through an explicitly linked fresh same-target
-  recovery launch; and
+  recovery launch, and never starts on a dirty tree: uncommitted work a lost
+  child left behind is first captured into a digest-verified artifact,
+  restored out of the worktree, handed forward as `recovered_patch`, and
+  committed by the continuation as its first action, while unverified or
+  unsupported dirt still blocks; and
 - a **successful repair** produces one immutable append-only recovery commit.
 
 This recovery is not accepted-launch fallback. After a launch is accepted,

package/assets/docs/workflows/skills/explainer-kit.md

@@ -434,11 +434,14 @@ typed adapter settings.
 Private integrations use the core boundary directly: resolve private inputs
 before the run, construct one versioned request, invoke the core once, then
 publish or link the versioned manifest after the run. Wrapper acceptance reads
-the complete post-run `publish-receipt/v2`, verifies every manifest artifact
-and the core run ID, and rejects foreign or stale receipts; `publish-receipt/v1`
-remains readable for replay of older runs only. Presets, private source
-systems, external-document synchronization, and personal destinations remain
-wrapper-owned.
+the complete post-run `publish-receipt/v2`, requires complete manifest and
+catalog evidence, verifies the core run ID, and rejects foreign or stale
+receipts; `publish-receipt/v1` remains readable for replay of older runs only.
+Public and protected v2 receipts must each provide exactly one entry for every
+finalized manifest artifact and exactly one generated catalog entry. The
+`explainer-kit` wrapper extension contract owns this rule; this page restates
+it for integrators. Presets, private source systems, external-document
+synchronization, and personal destinations remain wrapper-owned.
 
 V1 exposes no plugin registry or private mid-pipeline hook. Unsupported
 contract majors and identity mismatches fail closed instead of being guessed

package/assets/public-package-versions.json

@@ -1,6 +1,6 @@
 {
-  "cli": "0.2.56",
-  "docs-config": "0.2.56",
-  "docs-theme": "0.2.56",
-  "docs-transforms": "0.2.56"
+  "cli": "0.2.57",
+  "docs-config": "0.2.57",
+  "docs-theme": "0.2.57",
+  "docs-transforms": "0.2.57"
 }

package/assets/skills/analyze/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: analyze
-version: 0.1.1
+version: 0.2.0
 description: Multi-angle analysis of existing artifacts, codebases, documents, or systems. Examines what you have from six analysis angles and produces structured findings with prioritized recommendations.
 argument-hint: 'target [--context path]'
 user-invocable: true
@@ -49,15 +49,16 @@ Print this banner once at start:
 
 Then print step indicators before beginning work:
 
-- `[1/9] Parsing arguments…`
-- `[2/9] Reading context…` (if applicable)
-- `[3/9] Classifying input type…`
-- `[4/9] Selecting analysis angles…`
-- `[5/9] Checking sub-agent availability…`
-- `[6/9] Executing analysis angles…`
-- `[7/9] Checking for comparisons…` (if applicable)
-- `[8/9] Cross-angle synthesis…`
-- `[9/9] Writing artifact…`
+- `[1/10] Parsing arguments…`
+- `[2/10] Reading context…` (if applicable)
+- `[3/10] Classifying input type…`
+- `[4/10] Selecting analysis angles…`
+- `[5/10] Checking sub-agent availability…`
+- `[6/10] Executing analysis angles…`
+- `[7/10] Checking for comparisons…` (if applicable)
+- `[8/10] Cross-angle synthesis…`
+- `[9/10] Resolving output destination…`
+- `[10/10] Writing artifact…`
 
 For long-running operations, print a start line and a completion line:
 
@@ -72,7 +73,7 @@ Keep it concise; don't print a line for every shell command.
 
 ### Step 1: Parse arguments
 
-`[1/9] Parsing arguments…`
+`[1/10] Parsing arguments…`
 
 Parse from `$ARGUMENTS`:
 
@@ -91,7 +92,7 @@ The prompt should ask: _"What would you like me to analyze? Provide a file path,
 
 ### Step 2: Read context (if provided)
 
-`[2/9] Reading context…`
+`[2/10] Reading context…`
 
 If `--context` is specified:
 
@@ -107,7 +108,7 @@ If `--context` is not specified, skip this step.
 
 ### Step 3: Input type classification
 
-`[3/9] Classifying input type…`
+`[3/10] Classifying input type…`
 
 Classify the target to determine analysis angle emphasis:
 
@@ -126,7 +127,7 @@ Log: `→ Input type: {type}`
 
 ### Step 4: Analysis angle selection
 
-`[4/9] Selecting analysis angles…`
+`[4/10] Selecting analysis angles…`
 
 All six angles are ALWAYS run. The input type determines emphasis weighting (primary angles get deeper investigation):
 
@@ -157,7 +158,7 @@ Log the selected emphasis: `→ Primary: {angles} · Secondary: {angles}`
 ### Step 5: Sub-agent availability probe
 
 ```
-[5/9] Checking sub-agent availability…
+[5/10] Checking sub-agent availability…
   → analysis workers: {available | not resolved} ({reason})
   → Selected: Execution Tier {1|2|3} — {description}
 ```
@@ -184,7 +185,7 @@ Log the selected emphasis: `→ Primary: {angles} · Secondary: {angles}`
 
 ### Step 6: Execute analysis angles
 
-`[6/9] Executing analysis angles…`
+`[6/10] Executing analysis angles…`
 
 For each angle, the worker (sub-agent or self) receives:
 
@@ -209,7 +210,7 @@ Print progress for each angle:
 
 ### Step 7: Conditional /compare dispatch
 
-`[7/9] Checking for comparisons…`
+`[7/10] Checking for comparisons…`
 
 If any angle surfaces comparables (e.g. "this component could use library A or B"):
 

package/assets/skills/oat-brainstorm/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: oat-brainstorm
-version: 1.3.4
+version: 1.3.5
 description: Use when the user explicitly invokes the `brainstorm` verb, including `/oat-brainstorm`, "let's brainstorm", "brainstorm this", "can we brainstorm X", or "help me brainstorm X". For ambiguous exploratory phrasing ("I've been thinking", "what if", "help me think through"), do NOT auto-enter; respond conversationally and offer mode only after ≥2 sustained exploratory turns. Do NOT use for review, debug, PR, status, implementation, or active-workflow questions.
 disable-model-invocation: false
 user-invocable: true
@@ -191,7 +191,7 @@ Classify the opening brainstorm topic before running the Node preflight or print
 command -v node >/dev/null 2>&1 && echo "available" || echo "missing"
 ```
 
-- If `node` is **missing**: skip the offer entirely. Do not print the offer message. Log a one-line note in the conversation that the visual companion is unavailable in this environment (a state `oat-doctor` can pick up later: "visual companion suppressed — node not on PATH"). Proceed with `VISUAL_COMPANION = "unavailable"`.
+- If `node` is **missing**: skip the offer entirely. Do not print the offer message. Log a one-line note in the conversation for this session only: "visual companion suppressed — node not on PATH". Nothing persists that note, so no later diagnostic run can report it. Proceed with `VISUAL_COMPANION = "unavailable"`.
 - If `node` is **available**: print the offer as its own message. Suggested wording (adapt freely; the constraint is "own message, no other content"):
 
   > "Some of what we're working on might be easier to explain if I can show it in a local web browser — mockups, diagrams, side-by-side comparisons. The visual companion is bundled with this skill (a small Node-based local server). Want me to start it? (Requires opening a `localhost` URL.)"

package/assets/skills/oat-doctor/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: oat-doctor
-version: 1.1.0
+version: 1.2.0
 description: Use when you need to diagnose your OAT setup, check for outdated skills, identify misconfigurations, or get a summary of installed tools and config. Runs health checks and recommends corrective actions.
 argument-hint: '[--summary]'
 disable-model-invocation: true
@@ -151,47 +151,60 @@ For summary mode, compare installed skills against the full manifest to identify
 
 **Bundled skill manifest (source of truth):**
 
-Core pack skills:
+Each heading below names a pack exactly as `oat tools install <pack>` expects
+it, and lists every skill that pack ships.
 
-- oat-docs, oat-doctor
+`core` pack skills:
 
-Workflow pack skills:
+- oat-docs, oat-doctor
 
-- oat-project-capture, oat-project-clear-active, oat-project-complete
-- oat-project-design, oat-project-discover, oat-project-document
-- oat-project-implement, oat-project-import-plan, oat-project-new
-- oat-project-open, oat-project-plan, oat-project-plan-writing
-- oat-project-pr-final, oat-project-pr-progress, oat-project-progress
-- oat-project-promote-spec-driven, oat-project-quick-start
-- oat-project-reconcile, oat-project-revise, oat-project-review-provide
-- oat-project-review-receive, oat-project-review-receive-remote
-- oat-project-spec, oat-project-summary
-- oat-repo-knowledge-index, oat-worktree-bootstrap, oat-worktree-bootstrap-auto
-
-Ideas pack skills:
+`ideas` pack skills:
 
 - oat-idea-new, oat-idea-ideate, oat-idea-summarize, oat-idea-scratchpad
 
-Docs pack skills:
+`docs` pack skills:
+
+- authoring-docs, oat-agent-instructions-analyze
+- oat-agent-instructions-apply, oat-docs-analyze
+- oat-docs-apply, oat-docs-authoring, oat-docs-bootstrap
 
-- oat-agent-instructions-analyze, oat-agent-instructions-apply
-- oat-docs-analyze, oat-docs-apply
+`workflows` pack skills:
 
-Utility pack skills:
+- oat-cursor-cloud-projects, oat-explainer-kit, oat-project-autonomous
+- oat-project-capture, oat-project-clear-active, oat-project-complete
+- oat-project-design, oat-project-dispatch-subagents, oat-project-discover
+- oat-project-document, oat-project-implement, oat-project-import-plan
+- oat-project-new, oat-project-next, oat-project-open
+- oat-project-plan, oat-project-plan-writing, oat-project-pr-final
+- oat-project-pr-progress, oat-project-progress, oat-project-promote-spec-driven
+- oat-project-quick-start, oat-project-reconcile, oat-project-retro
+- oat-project-retro-file, oat-project-revise, oat-project-review-provide
+- oat-project-review-provide-remote, oat-project-review-receive
+- oat-project-review-receive-remote, oat-project-spec, oat-project-split
+- oat-project-summary, oat-repo-knowledge-index, oat-worktree-bootstrap
+- oat-worktree-bootstrap-auto, oat-wave-execute, oat-wave-program
+- oat-wrap-up
+
+`utility` pack skills:
 
-- create-agnostic-skill
-- oat-repo-maintainability-review, oat-review-provide
-- oat-review-receive, oat-review-receive-remote
+- create-agnostic-skill, explainer-kit, oat-dispatch-subagents
+- subagent-orchestration, oat-repo-improve, oat-repo-maintainability-review
+- oat-review-provide, oat-review-provide-remote, oat-review-receive
+- oat-review-receive-remote
 
-Project management pack skills:
+`project-management` pack skills:
 
-- oat-pjm-add-backlog-item, oat-pjm-update-repo-reference
-- oat-pjm-review-backlog
+- oat-pjm-add-backlog-item, oat-pjm-decision
+- oat-pjm-update-repo-reference, oat-pjm-review-backlog
 
-Research pack skills:
+`research` pack skills:
 
 - analyze, compare, deep-research
-- skeptic, synthesize
+- recon, skeptic, synthesize
+
+`brainstorm` pack skills:
+
+- oat-brainstorm
 
 For each pack, determine:
 
@@ -224,12 +237,12 @@ OAT ▸ DOCTOR SUMMARY
 | Pack      | Scope   | Skills | Status   |
 | --------- | ------- | ------ | -------- |
 | core      | user    | 2/2    | current  |
-| docs      | project | 4/4    | current  |
-| workflows | project | 26/26  | current  |
+| docs      | project | 7/7    | current  |
+| workflows | project | 39/39  | current  |
 | ideas     | user    | 4/4    | current  |
-| project-management | project | 3/3 | current |
-| research  | project | 5/5    | current  |
-| utility   | project | 5/5    | outdated |
+| project-management | project | 4/4 | current |
+| research  | project | 6/6    | current  |
+| utility   | project | 10/10  | outdated |
 
 ## Outdated Skills
 
@@ -239,11 +252,11 @@ OAT ▸ DOCTOR SUMMARY
 
 ## Available But Not Installed
 
-- **docs** pack: oat-docs-analyze, oat-docs-apply, oat-agent-instructions-analyze, oat-agent-instructions-apply (4 skills available)
+- **docs** pack: authoring-docs, oat-agent-instructions-analyze, oat-agent-instructions-apply, oat-docs-analyze, oat-docs-apply, oat-docs-authoring, oat-docs-bootstrap (7 skills available)
   → Run: oat tools install docs --scope {scope}
-- **project-management** pack: oat-pjm-add-backlog-item, oat-pjm-update-repo-reference, oat-pjm-review-backlog (3 skills available)
+- **project-management** pack: oat-pjm-add-backlog-item, oat-pjm-decision, oat-pjm-update-repo-reference, oat-pjm-review-backlog (4 skills available)
   → Run: oat tools install project-management --scope {scope}
-- **research** pack: analyze, compare, deep-research, skeptic, synthesize (5 skills available)
+- **research** pack: analyze, compare, deep-research, recon, skeptic, synthesize (6 skills available)
   → Run: oat tools install research --scope {scope}
 
 ## Configuration

package/assets/skills/oat-idea-summarize/SKILL.md

@@ -1,11 +1,11 @@
 ---
 name: oat-idea-summarize
-version: 1.2.1
+version: 1.3.0
 description: Use when an idea is mature enough to move from brainstorming into the backlog. Generates a summary document and adds the idea to the backlog.
 argument-hint: '[--global]'
 disable-model-invocation: true
 user-invocable: true
-allowed-tools: Read, Write, Grep, AskUserQuestion
+allowed-tools: Read, Write, Bash, Glob, Grep, AskUserQuestion
 ---
 
 # Summarize Idea

package/assets/skills/oat-project-autonomous/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: oat-project-autonomous
-version: 1.0.9
+version: 1.0.10
 description: Use when a user explicitly asks to run an OAT project autonomously end-to-end. Activates session-only autonomy, resumes the correct lifecycle phase, and drives the existing OAT skills through final PR or a reported boundary.
 argument-hint: '<goal | project-slug | ticket-ref>'
 disable-model-invocation: true
@@ -174,8 +174,8 @@ repository-policy boundary.
 ### Step 0.5: Capability Detection and Tier Selection
 
 Before artifact writes, external side effects, tests, or long-running work,
-load `oat-project-dispatch-subagents`, which in turn requires
-`oat-dispatch-subagents`. Probe the roles and dispatch surfaces needed for
+load the current `oat-project-dispatch-subagents/SKILL.md` and follow it, which
+in turn requires `oat-dispatch-subagents`. Probe the roles and dispatch surfaces needed for
 lifecycle workers and independent reviewers.
 
 Classify each required capability as:
@@ -247,7 +247,9 @@ If evidence cannot support the choice without changing product intent, stop at
 a product-judgment boundary.
 
 Invoke `oat-project-new` for spec-driven mode or `oat-project-quick-start` for
-quick mode. Existing projects retain their persisted workflow mode.
+quick mode: load the selected creation skill's current `SKILL.md` and follow it,
+or dispatch a child that carries it, never a remembered outcome. Existing
+projects retain their persisted workflow mode.
 
 ### Step 2.5: Persist Autonomous Explainer Intent
 
@@ -341,8 +343,11 @@ fi
 ### Step 5: Invoke Lifecycle Skills and Reviews
 
 Invoke each lifecycle skill by name and let it own its complete workflow,
-artifacts, gates, commits, and state transitions. Re-read project status after
-each return and route to the next earliest incomplete owner.
+artifacts, gates, commits, and state transitions. Invoking a lifecycle skill
+means loading that skill's current `SKILL.md` and following it, or dispatching a
+child that carries it; a remembered outcome or ambient discovery is not
+compliant. Re-read project status after each return and route to the next
+earliest incomplete owner.
 
 When planning finds an incomplete dispatch ladder, apply the gate inventory's
 autonomous ownership resolution instead of treating ordinary non-interactive
@@ -359,8 +364,9 @@ Block when the ladder remains incomplete after adoption.
 
 At every required artifact or code review:
 
-1. Resolve the route before launch through
-   `oat-project-dispatch-subagents` and its generic dispatch substrate.
+1. Resolve the route before launch by loading the current
+   `oat-project-dispatch-subagents/SKILL.md`, following it, and using its
+   generic dispatch substrate.
 2. Prefer a configured independent gate route when its exact target is
    available; otherwise select a policy-compliant cross-family reviewer.
 3. Select same-family/context-independent review only when no second family is
@@ -406,8 +412,9 @@ Ensure `oat-project-implement` resolves its immutable
   credential stops at its boundary.
 
 The orchestrator's responsibility is to keep invoking
-`oat-project-implement` until that sequence reaches `complete` or reports a
-boundary. Never wrap or replace implement's phase-agent topology.
+`oat-project-implement` — loading the current `oat-project-implement/SKILL.md`
+and following it on every resume — until that sequence reaches `complete` or
+reports a boundary. Never wrap or replace implement's phase-agent topology.
 
 ### Step 8: Commit and Push Phase Boundaries