adeelahmad · Oct 18, 2026 · Oct 18, 2026 · Oct 18, 2026 · Oct 18, 2026 · Oct 18, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -50,6 +50,14 @@ tag in lockstep. Tag releases as `vMAJOR.MINOR.PATCH`.
   overrides it (`breach:` line) and acts per class (SKILL.md "Breach classes").
 
 ### Added
+- Stuck release: a worker that hits the same gate block 3 times in a row (one gate, task,
+  attempt and worktree; `AGENTIC_STUCK_BLOCKS`) is released **unverified** instead of
+  looping until its time box (live run 10 lost 29 minutes to a RED worker that stopped 29
+  times on one unchanged reason). The release is not a pass: the gate leaves
+  `.agentic/state/stuck/<task>.<attempt>` with the reason, `budget watch` prints a `STUCK`
+  line once, and the supervisor re-dispatches with feedback or re-plans. A different reason
+  or a new attempt resets the count; a self-check never releases. Every gate verdict now
+  logs its worktree and a checksum of the block reason. 7 conformance checks (§6.9).
 - **Portable project config** (ROADMAP §4.9). `.aladdin/config.yaml` (budgets, context,
   semantic, models, provenance, memory, git, plus per-role budget profiles) replaces
   `docs/agents/defaults.md` as the human's file. `aloop config migrate` writes it from the
@@ -237,6 +245,28 @@ tag in lockstep. Tag releases as `vMAJOR.MINOR.PATCH`.
   reading it stays allowed. 2 checks failed before.
 
 ### Fixed
+- `gate-validate-artifact`'s block now carries md-db's diagnostics (the missing section or field
+  and its hint): a PostToolUse block hands the model only stderr, and the bare "failed the
+  schema" made the planner guess (live e2e run 11: every planning artifact was written twice,
+  12 of 25 gate verdicts).
+- `gate-dispatch` refuses a task contract with no `STORY_DIR`, or one naming a dir that does
+  not exist, and tells a whole-wave structural review to use the sprint dir
+  `docs/agents/sprintN/` (run 11: the reviewer's contract had none, 2 blocks the worker could
+  not fix). 5 new dispatch checks.
+- OpenCode: a stuck release (exit 0 with `STUCK … NOT verified`) is reported to the supervisor
+  as "released UNVERIFIED, do not merge", never as a silent pass; 2 new checks.
+- `make test-hosts` masked a failing OpenCode test as "node not found".
+- The host tests (Codex hooks, OpenCode plugin, rename-compat, budget, fail-closed, scope,
+  dispatch) each leaked a temp dir, and the Codex and OpenCode ones started a real background
+  cargo build over their fake tools (~500MB per run; 57 leaked dirs filled the disk and killed
+  e2e run 11 with ENOSPC). They now remove their temp dir on exit and set
+  `AGENTIC_NO_AUTOINSTALL=1`; `e2e-live` drops its cargo target dirs after the run.
+- The comms gate's "no output.md" block now says HOW to append the report: a shell heredoc
+  from the worktree, because Claude Code's Write/Edit refuse a path outside it and
+  `docs/agents/` lives only in the main tree (e2e run 10: a RED worker stopped 29 times on
+  it). The worker personas say the same.
+- A worktree `task.env` copy OLDER than the supervisor's contract (a re-dispatch bumped
+  `ATTEMPT`) is stale and is ignored, not reported as tampering (e2e run 10: 3 blocks).
 - `ensure-tools` rebuilds the bundled md-db / ctx-symbols when their sources change, not
   only on a version bump; e2e run 9 ran a stale ctx-symbols that still had the duplicate bug.
 - `gate-stage2-complete` names the expected `sN-NN-<slug>/` story dir (and a misnamed one it

diff --git a/Makefile b/Makefile
@@ -113,7 +113,7 @@ test-hosts: ## Budget/worktree regression + Codex hooks emulation + OpenCode plu
 	python3 scripts/test/dispatch.test.py
 	python3 scripts/test/ensure-tools.test.py
 	python3 scripts/test/fail-closed.test.py
-	@command -v node >/dev/null && node scripts/test/opencode-plugin.test.mjs || { echo "node not found — skipped OpenCode test (CI runs it)"; }
+	@if command -v node >/dev/null; then node scripts/test/opencode-plugin.test.mjs; else echo "node not found — skipped OpenCode test (CI runs it)"; fi
 
 .PHONY: conformance
 conformance: ## ROADMAP §6 conformance fixtures on today's gates (needs cargo; no network, no tokens)

diff --git a/docs/OVERVIEW.md b/docs/OVERVIEW.md
@@ -95,9 +95,16 @@ The model can ignore a skill; it cannot ignore an exit code.
 **Working and tested:**
 - The full planning → execution loop on Claude Code; Codex and OpenCode via generated
   agents and the same gates.
-- CI on every PR: Rust fmt/clippy/tests for both backends, shellcheck on every gate, Codex
-  hooks emulation + OpenCode end-to-end (18 checks), `claude plugin validate --strict`,
-  eval-suite and version-lockstep checks.
+- CI on every PR: Rust fmt/clippy/tests for both backends, shellcheck on every gate, the
+  ROADMAP §6 conformance fixtures (real gates on a fixture crate, 54 checks), the host tests
+  (budget/worktree, Codex hooks emulation, supervisor scope, hook matchers, dispatch,
+  ensure-tools, fail-closed: 73 checks; OpenCode end-to-end: 20), `claude plugin validate
+  --strict`, eval-suite and version-lockstep checks.
+- The north-star run, `make e2e-live YES=1 [INTENT=wordfreq]`: a one-line intent becomes
+  planned, test-first, gate-verified, merged code in a real Claude Code session with the
+  approval given up front. Two intents pass every check (semver parser: 11/11, ~$0.65–1.15,
+  11–19 min; word frequencies with two dependent stories: 12/12, $0.86–1.56, 13–45 min). Every run so
+  far has found a gate false positive that unit tests missed; ROADMAP M0.8 lists them.
 - One measured real run (*TheNoteBook*, a Go backend): ~17K LOC, 322 tests, 74 commits in 6
   sprints for ~1.55M output tokens, 76% of it on Sonnet. See the
   [README](../README.md#token-efficiency-a-measured-run).

diff --git a/docs/ROADMAP.md b/docs/ROADMAP.md
@@ -14,11 +14,11 @@ spec; this table is the plan. Versions are targets, not promises.
 | Milestone | Target | Contents | Status |
 |---|---|---|---|
 | **M0 — Hygiene** | v0.10.x | Mark `DESIGN` / `ARCHITECTURE` / `KICKSTART` historical; fix stale counts (8 → 9 roles) and the README CI section; add the `CONTRIBUTING.md` the README badge links to; one live-session smoke test of the hook wiring (open since `DEVLOG` v0.1) | done: live smoke test `make smoke-live YES=1` (6 checks, ~$0.004 on haiku); its first run found the Linux fail-open in `gate-supervisor-scope` (fixed, regression test added) |
-| **M0.8 — North-star run** | v0.11.x | `make e2e-live YES=1`: a one-line intent (semver parser) goes through retrospective, intake, standards, planner, RED/SCAFFOLD/GREEN per task, structural review and the final gate to merged, tested code in a real Claude Code session, with the approval given up front. Reason: nothing had ever run the whole loop live; the first run found that no role gate fired on plugin installs | run 1: done end to end, 19 min, $1.15 (sonnet supervisor, haiku workers), 22 tests on main, 9/11 checks; found and fixed the `aloop:<role>` matcher bug, a selfcheck crash and copy-merging. Run 2: **11/11**, 12m46s, $0.92, host gates on every role, worktree branches merged with `--no-ff`; found gates looping on the supervisor's contract slips (fixed). Run 3: 11/11, 12m00s, **$0.79**; found GREEN looping on RED's unformatted tests (fixed: RED/SCAFFOLD run the format check) and selfcheck missing the contract in Claude Code worktrees (fixed: the contract is copied in). Run 4 (all fixes): 11/11, 17m56s, $1.10, 12 gate blocks and no loops (28 → 17 → 12). Cost varies $0.79–1.15 per run; next: cut it (scaffold symbol parsing slips such as `(enum,`, supervisor turns), then a second intent beyond the semver fixture. Run 5: 11/11 but 29 min, $1.83, 44 blocks (the planner made 5 tasks; 27 false isolation blocks after workers `cd`'d to the main tree, fixed; subagents also read the plugin's own source, which costs tokens). Run 6: 11/11, **11m10s, $0.65**, plugin-source reads 30 → 2 after the agent prompts said not to; found a dispatch-before-contract loop (25 blocks, fixed by `gate-dispatch`) and annotated scaffold-symbol lines (fixed). Run 7 (after #9): 11/11, 13m46s, $0.98, 20 blocks, no dispatch loop; found RED blocked for listing an earlier task's passing tests, scaffold-symbols tables with a header row and trait-impl lines, and a guessed memory.md frontmatter (all fixed; `aloop init` now seeds memory.md). The re-plan coverage loss is now caught by `gate-final` (every planned test must be on main). Run 8, **second intent** (`INTENT=wordfreq`, two dependent stories): 12/12, 18m05s, $1.53, 54 blocks; found scope globs expanded against the worktree (29 GREEN blocks) and `mod common;` across test crates reported as a HIGH duplicate (57 review halts), both fixed. Run 9 (wordfreq again): 12/12, 19m42s, $1.56, 29 blocks; the ctx-symbols fix had not reached the run (`ensure-tools` rebuilt only on a version bump, fixed) and the planner named story dirs `S1-01/` (block message now names `s1-01-<slug>/`). Next: rerun wordfreq with the rebuilt tools, then supervisor-written GREEN instructions overreaching the next task |
+| **M0.8 — North-star run** | v0.11.x | `make e2e-live YES=1`: a one-line intent (semver parser) goes through retrospective, intake, standards, planner, RED/SCAFFOLD/GREEN per task, structural review and the final gate to merged, tested code in a real Claude Code session, with the approval given up front. Reason: nothing had ever run the whole loop live; the first run found that no role gate fired on plugin installs | run 1: done end to end, 19 min, $1.15 (sonnet supervisor, haiku workers), 22 tests on main, 9/11 checks; found and fixed the `aloop:<role>` matcher bug, a selfcheck crash and copy-merging. Run 2: **11/11**, 12m46s, $0.92, host gates on every role, worktree branches merged with `--no-ff`; found gates looping on the supervisor's contract slips (fixed). Run 3: 11/11, 12m00s, **$0.79**; found GREEN looping on RED's unformatted tests (fixed: RED/SCAFFOLD run the format check) and selfcheck missing the contract in Claude Code worktrees (fixed: the contract is copied in). Run 4 (all fixes): 11/11, 17m56s, $1.10, 12 gate blocks and no loops (28 → 17 → 12). Cost varies $0.79–1.15 per run; next: cut it (scaffold symbol parsing slips such as `(enum,`, supervisor turns), then a second intent beyond the semver fixture. Run 5: 11/11 but 29 min, $1.83, 44 blocks (the planner made 5 tasks; 27 false isolation blocks after workers `cd`'d to the main tree, fixed; subagents also read the plugin's own source, which costs tokens). Run 6: 11/11, **11m10s, $0.65**, plugin-source reads 30 → 2 after the agent prompts said not to; found a dispatch-before-contract loop (25 blocks, fixed by `gate-dispatch`) and annotated scaffold-symbol lines (fixed). Run 7 (after #9): 11/11, 13m46s, $0.98, 20 blocks, no dispatch loop; found RED blocked for listing an earlier task's passing tests, scaffold-symbols tables with a header row and trait-impl lines, and a guessed memory.md frontmatter (all fixed; `aloop init` now seeds memory.md). The re-plan coverage loss is now caught by `gate-final` (every planned test must be on main). Run 8, **second intent** (`INTENT=wordfreq`, two dependent stories): 12/12, 18m05s, $1.53, 54 blocks; found scope globs expanded against the worktree (29 GREEN blocks) and `mod common;` across test crates reported as a HIGH duplicate (57 review halts), both fixed. Run 9 (wordfreq again): 12/12, 19m42s, $1.56, 29 blocks; the ctx-symbols fix had not reached the run (`ensure-tools` rebuilt only on a version bump, fixed) and the planner named story dirs `S1-01/` (block message now names `s1-01-<slug>/`). Run 10 (rebuilt tools): 12/12, $1.20, but 44m48s and 45 blocks: the duplicate false positive is gone (0 HIGH); a RED worker's Write/Edit to the story's output.md was refused by Claude Code's worktree isolation and it stopped 29 times without trying a shell append (fixed: the block and the personas say how), plus 3 stale-`ATTEMPT` tamper blocks (fixed). Fixed after run 10: a worker that repeats an identical block 3 times is released unverified with a `STUCK` marker the supervisor acts on (reason: 29 minutes lost in run 10). Run 11 (all run-10 fixes): 12/12, **12m56s, $0.86, 6 blocks** (from 44m48s, $1.20, 45): no r
8BD0
epeated block, one RED report-append block fixed on the first retry, one scaffold fmt block, two structural-reviewer comms blocks (STORY_DIR unset for a sprint-wide review, then a stale block) and 12 `gate-validate-artifact` write blocks on the planner's artifacts. Fixed after run 11: the artifact-write block carries md-db's diagnostics (the planner guessed twice per artifact: 12 of 25 verdicts), `gate-dispatch` requires an existing `STORY_DIR` and names the sprint dir for a whole-wave structural review (2 blocks), and OpenCode reports a stuck release as unverified. Next: rerun wordfreq to measure, then supervisor-written GREEN instructions overreaching the next task |
 | **M0.6 — Gates fail closed** | v0.11.x | A gate that crashes (exit 1, e.g. an unbound variable) is a non-blocking hook error on Claude Code, so the action goes through. Make every gate trap unexpected errors into exit 2 with the reason, and add a conformance case per gate. Reason: the smoke run showed one real instance | done: `_failclosed.sh` on all 10 sub-agent gates + supervisor-scope inside a sprint; `scripts/test/fail-closed.test.py` (5 checks). Left fail-open on purpose: `gate-commit-author` and `gate-validate-artifact` outside a sprint, which see every session on the machine |
-| **M0.7 — Supervisor-owned task contract** | v0.11.x | The contract lives in the main tree, `.agentic/state/tasks/<TASK_ID>.env`, and the dispatch prompt's first line names the task; gates and the time box find it from the host's transcript, a worktree `task.env` that disagrees blocks, and sub-agents may not Write/Edit `.agentic/state/` or `task.env`. Settled by a live probe: Claude Code's `isolation: "worktree"` makes `.claude/worktrees/agent-<id>` at dispatch, so a pre-written in-worktree contract was impossible there. Bash writes are guarded too (a command naming the contract with a write verb is blocked; a heuristic, so an obfuscated write still gets through). On hosts with no sub-agent transcript the worktree's `TASK_ID` (or its `.agentic/worktrees/<id>` path) picks the contract | done: `make conformance` (30) + `scripts/test/{budget-worktree,supervisor-scope}.test.py` |
+| **M0.7 — Supervisor-owned task contract** | v0.11.x | The contract lives in the main tree, `.agentic/state/tasks/<TASK_ID>.env`, and the dispatch prompt's first line names the task; gates and the time box find it from the host's transcript, a worktree `task.env` that disagrees blocks, and sub-agents may not Write/Edit `.agentic/state/` or `task.env`. Settled by a live probe: Claude Code's `isolation: "worktree"` makes `.claude/worktrees/agent-<id>` at dispatch, so a pre-written in-worktree contract was impossible there. Bash writes are guarded too (a command naming the contract with a write verb is blocked; a heuristic, so an obfuscated write still gets through). On hosts with no sub-agent transcript the worktree's `TASK_ID` (or its `.agentic/worktrees/<id>` path) picks the contract | done: `make conformance` (54 today) + `scripts/test/{budget-worktree,supervisor-scope}.test.py` |
 | **M0.5 — Rename to aloop** | v0.11.0 | Plugin, marketplace, skill, CLI (`aloop`, `aloop init`), OpenCode plugin file and docs renamed; `agentic`, `agentic init`, `agentic-init` and `agentic bind` kept as aliases; OpenCode/Codex installs clean up pre-rename files. Data layout and settings keys unchanged (moved to M1). GitHub repo rename is the owner's call | done: `scripts/test/rename-compat.test.py` |
-| **M1 — Trustworthy state** | v0.12 | Session 1: conformance fixtures around current behaviour, canonical IDs and the reference remap (§4.5); then move the data layout to the new name (`.agentic/` → `.aloop/`, `AGENTIC_*` → `ALOOP_*`) with a read-old fallback, proven by those fixtures | in progress: §6.2–6.9 fixtures pass on today's gates (`make conformance`, 30 checks incl. task.env tampering and the main-tree contract; found and fixed the whole-directory schema-gate bug); next: canonical IDs and the reference remap (§4.5) |
+| **M1 — Trustworthy state** | v0.12 | Session 1: conformance fixtures around current behaviour, canonical IDs and the reference remap (§4.5); then move the data layout to the new name (`.agentic/` → `.aloop/`, `AGENTIC_*` → `ALOOP_*`) with a read-old fallback, proven by those fixtures | in progress: §6.2–6.9 fixtures pass on today's gates (`make conformance`, 54 checks incl. task.env tampering, the main-tree contract and every false positive the live runs found; found and fixed the whole-directory schema-gate bug); next: canonical IDs and the reference remap (§4.5) |
 | **M2 — Budgets that plan** | v0.13 | Sessions 2 and 4: budget semantics (§3.1–3.5), sprint envelope and admission control (§4.2), breach classes, failure envelope, grace rule, lesson candidates (§4.3) | not started |
 | **M3 — Clean boundaries** | v0.14 | Session 3: canonical handoff, rotation and the resume rule (§4.4) | not started |
 | **M4 — Intent in, autonomy optional** | v0.15 | Session 5: Intent Contract, clarification policy, autonomous mode (§4.1) | not started |

diff --git a/plugin/agents/green-worker.md b/plugin/agents/green-worker.md
@@ -37,7 +37,8 @@ your init block's `### Budget` line has the exact numbers). Plan for it:
 Your brief is the LAST `## … · green-worker · …` block in `$STORY_DIR/init.md`; read the
 whole file plus the red-worker and scaffolder blocks in `$STORY_DIR/output.md` (what was
 tested, what was stubbed). When done you MUST **append** your report as a new block to
-`$STORY_DIR/output.md` — never rewrite an earlier block:
+`$STORY_DIR/output.md` — never rewrite an earlier block (append with a shell heredoc, `cat >> "$STORY_DIR/output.md" <<'EOF'`: the Write/Edit
+tools refuse a path outside your worktree, and docs/agents/ lives only in the main tree):

     ## <task> · attempt <N> · green-worker · <ISO8601>
     status: ok            (or retry|escalate)

diff --git a/plugin/agents/red-worker.md b/plugin/agents/red-worker.md
@@ -37,7 +37,8 @@ your init block's `### Budget` line has the exact numbers). Plan for it:
 Your brief is the LAST `## … · red-worker · …` block in `$STORY_DIR/init.md`; read the
 whole file for prior context (earlier feedback, the chain so far). When done you MUST
 **append** your report as a new block to `$STORY_DIR/output.md` — create it with a
-`type: output` frontmatter if absent, and NEVER rewrite an earlier block:
+`type: output` frontmatter if absent, and NEVER rewrite an earlier block (append with a shell heredoc, `cat >> "$STORY_DIR/output.md" <<'EOF'`: the Write/Edit
+tools refuse a path outside your worktree, and docs/agents/ lives only in the main tree):
 
     ## <task> · attempt <N> · red-worker · <ISO8601>
     status: ok            (or retry|escalate)

diff --git a/plugin/agents/scaffolder.md b/plugin/agents/scaffolder.md
@@ -43,7 +43,8 @@ GREEN. Scaffolding is safe to repeat.
 Your brief is the LAST `## … · scaffolder · …` block in `$STORY_DIR/init.md`; read the
 red-worker block in `$STORY_DIR/output.md` for the symbols the tests reference. When done
 you MUST **append** your report as a new block to `$STORY_DIR/output.md` — never rewrite
-an earlier block:
+an earlier block (append with a shell heredoc, `cat >> "$STORY_DIR/output.md" <<'EOF'`: the Write/Edit
+tools refuse a path outside your worktree, and docs/agents/ lives only in the main tree):
 
     ## <task> · attempt <N> · scaffolder · <ISO8601>
     status: ok

diff --git a/plugin/agents/structural-reviewer.md b/plugin/agents/structural-reviewer.md
@@ -26,9 +26,11 @@ Everything is on PATH — never search for it: `selfcheck`, `md-db`, `ctx-symbol
 - Read-only. Writes only output.md findings; fixes nothing itself.
 
 ## Comms — the channel (story-bound, append-only)
+`$STORY_DIR` is the story's dir, or the sprint dir (`docs/agents/sprintN/`) for a whole-wave review.
 Read the story's chain in `$STORY_DIR/output.md` (the red/scaffold/green blocks) plus the
 merged code. When done you MUST **append** your findings as a new block to
-`$STORY_DIR/output.md` — never rewrite an earlier block:
+`$STORY_DIR/output.md` — never rewrite an earlier block (append with a shell heredoc, `cat >> "$STORY_DIR/output.md" <<'EOF'`: the Write/Edit
+tools refuse a path outside your worktree, and docs/agents/ lives only in the main tree):
 
     ## <task> · attempt <N> · structural-reviewer · <ISO8601>
     status: ok            (or reject — foundation-poisoning)