adeelahmad · Oct 18, 2026 · Oct 18, 2026 · Oct 18, 2026 · Oct 18, 2026 · Oct 18, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -22,6 +22,13 @@ tag in lockstep. Tag releases as `vMAJOR.MINOR.PATCH`.
 
 ## [Unreleased]
 
+### Added
+- The planner's stop gate (`gate-stage2-complete` on SubagentStop) now logs its verdicts to `gates.jsonl` and the event stream like every other gate, and the stuck release covers the planner: the same block three times in a row releases its stop unverified with a `state/stuck/plan-sprintN.<agent>` marker the supervisor acts on (a re-dispatch is a new agent id and starts over). A direct call (the supervisor's own check, `selfcheck`) is logged as a self-check and never releases. Reason: live e2e run 19's re-planning planner was blocked 9 times on one unchanged reason, none counted. Test: `scripts/test/planner-stuck.test.py`.
+
+### Fixed
+- Planner persona: a later task's RED tests must fail against what the earlier tasks leave on main, not only the empty tree (live run 19: three of T2's tests passed against T1's placeholder, costing a re-plan).
+- `targets check` no longer rejects a `create` change whose file already exists **and defines its planned symbol**: that change has shipped. A mid-sprint re-plan re-runs the planner's stop gate over the whole plan, and live e2e run 19 blocked the planner 9 times on T1's merged `create tests/parse.rs` while its brief said not to touch T1. `target` now resolves such a change to `file:line`. Test: `scripts/test/semantic-targets.test.py`.
+
 ### Added
 - `gate-supervisor-scope` now also enforces **the supervisor never commits on main**: a `git commit` from the main agent in the main tree while a sprint is live is blocked (a commit inside a task worktree, the initial commit of an unborn repo, `task-worktree merge` and a stale lock pass). Reason: live e2e run 16's `Track Cargo.lock` commit on main traced to nothing; the rule was prose in SKILL.md only. Test: `scripts/test/supervisor-scope.test.py`.
 

diff --git a/Makefile b/Makefile
@@ -116,6 +116,7 @@ test-hosts: ## Budget/worktree regression + Codex hooks emulation + OpenCode plu
 	python3 scripts/test/codex-hooks.test.py
 	python3 scripts/test/rename-compat.test.py
 	python3 scripts/test/supervisor-scope.test.py
+	python3 scripts/test/planner-stuck.test.py
 	python3 scripts/test/hook-matchers.test.py
 	python3 scripts/test/dispatch.test.py
 	python3 scripts/test/log-execution.test.py

diff --git a/docs/ROADMAP.md b/docs/ROADMAP.md
diff --git a/docs/TRACKING.md b/docs/TRACKING.md
@@ -31,6 +31,7 @@ Status of every requirement in [ROADMAP.md](ROADMAP.md), in the four states §8
 | 4.7 | Decision rationale + authority record | IMPLEMENTED_TESTED | `scripts/test/events.test.py` |
 | 4.7 | Lineage: commit → intent → approver | IMPLEMENTED_TESTED | `scripts/test/commit-trace.test.py` |
 | 4.7 | Every commit on main is a task's: the supervisor's own `git commit` on main is blocked mid-sprint | IMPLEMENTED_TESTED | `scripts/test/supervisor-scope.test.py` |
+| 6.9 | Stuck release covers the planner; its stop verdicts are logged like every gate | IMPLEMENTED_TESTED | `scripts/test/planner-stuck.test.py` |
 | 4.7 | Commit messages from rationale; graph export | IMPLEMENTED_TESTED | `scripts/test/commit-trace.test.py` |
 | 4.8 | Cross-project impact provider seam | IMPLEMENTED_TESTED | `scripts/test/impact-provider.test.py` |
 | 4.9 | Usage reporting: classes apart, exact or marked, unreported classes named | IMPLEMENTED_TESTED | `scripts/test/budget-meter.test.py` |

diff --git a/plugin/agents/planner.md b/plugin/agents/planner.md
@@ -49,6 +49,14 @@ hard; 150k / 20 min soft — see `docs/agents/defaults.md`). A worker still runn
 the hard limit is STOPPED by the hook and the task comes back to you to split — never retried.
 So size each task so one worker can do its RED or its GREEN well inside the soft
 limit: a handful of tests, one or two files, one behaviour. When in doubt, split.
+A later task's RED tests must fail against what the EARLIER tasks leave on main, not only
+against the empty tree: `gate-red-verify` requires every new test to fail by assertion, so
+a task whose planned tests already pass against the previous task's skeleton (a placeholder
+that maps every non-success to one error variant, a default where a `todo!()` was) cannot be
+RED and costs a mid-sprint re-plan (live run 19: three of T2's tests passed against T1's
+placeholder). Plan each later task's tests around behaviour the earlier task's contract does
+not provide, and state in the earlier task's summary what its skeleton returns for inputs
+outside its scope, so the later tests can assert something it gets wrong.
 If defaults.md sets `AGENTIC_TIME_MODE=auto`, propose ONE stable task duration for the
 project (with a min and a max) to the supervisor, who records it with `timebox propose`;
 ask for a revision only with evidence (a stats.md, an event id).

diff --git a/plugin/bin/README.md b/plugin/bin/README.md
@@ -34,7 +34,7 @@ ctx-symbols) → WARN + grep fallback; never a silent pass.
 | gate-structural-integrity | SubagentStop · structural-reviewer | appended output.md block; orphan/parallel/duplicate via ctx-symbols | no/stale output.md block / foundation-poisoning finding |
 | gate-final                | SubagentStop · final-gate   | full matrix; zero suppressions; all plan-ready ticked | matrix red / suppression / unticked |
 | gate-standards-cited      | SubagentStop · standards    | standards.md citations resolve | dangling citation |
-| gate-stage2-complete      | supervisor self-check + SubagentStop · planner | every story Stage-2 valid; no TBW; sprint self-contained (no later-sprint story IDs, no work deferred to a later sprint) | incomplete or spilling planning |
+| gate-stage2-complete      | supervisor self-check + SubagentStop · planner | every story Stage-2 valid; no TBW; sprint self-contained (no later-sprint story IDs, no work deferred to a later sprint); verdicts logged like every gate, and the stuck release covers the planner (3 identical blocks → stop released unverified, marker `state/stuck/plan-sprintN.<agent>`) | incomplete or spilling planning |
 | gate-plan-shape           | called by gates             | each plan checkbox carries `path::fn` | malformed plan |
 | gate-ledger-format        | called by gates             | execution.log line shape | malformed ledger |
 

diff --git a/plugin/bin/_gatelib.sh b/plugin/bin/_gatelib.sh
@@ -76,7 +76,9 @@ _gate_stuck() {
   [ "$self" = false ] && [ -n "${TASK_ID:-}" ] && [ -n "${ATTEMPT:-}" ] || return 1
   # The build roles, plus the two read-only roles whose blocks can name a defect only the
   # supervisor can act on (live e2e run 12: a HIGH structural finding blocked the reviewer 10x).
-  case "${AGENT_ROLE:-}" in red-worker|scaffolder|green-worker|structural-reviewer|final-gate) ;; *) return 1 ;; esac
+  # And the planner (live e2e run 19: 9 identical blocks on a re-plan); its key is the sprint
+  # + the host's agent id (gate-stage2-complete), so a re-dispatch starts over.
+  case "${AGENT_ROLE:-}" in red-worker|scaffolder|green-worker|structural-reviewer|final-gate|planner) ;; *) return 1 ;; esac
   [ "$n" -ge 2 ] 2>/dev/null && [ -f "$log" ] || return 1
   prior="$(grep -F "\"gate\":\"${GATE_NAME:-?}\"" "$log" | grep -F "\"task\":\"$TASK_ID\",\"attempt\":\"$ATTEMPT\"" \
     | grep -F "\"wt\":\"$(repo_dir)\"" | grep -F '"selfcheck":false' | tail -n $((n - 1)) | grep -c -F "\"exit\":2,\"selfcheck\":false,\"reason\":\"$rh\"}")"
@@ -85,8 +87,14 @@ _gate_stuck() {
   printf 'role=%s gate=%s task=%s attempt=%s blocks=%s\n%s\n' "${AGENT_ROLE:-}" "${GATE_NAME:-?}" "$TASK_ID" "$ATTEMPT" "$n" \
     "$GATE_BLOCK_REASON" > "$AGENTIC_STATE/stuck/$TASK_ID.$ATTEMPT" 2>/dev/null
   echo "[${GATE_NAME:-gate}] STUCK — the same block $n times in a row; stop released, NOT verified." >&2
-  echo "  This attempt is not a pass. Supervisor: do not merge; discard the worktree and re-dispatch" >&2
-  echo "  $TASK_ID with a Feedback block that addresses the reason (.agentic/state/stuck/$TASK_ID.$ATTEMPT), or re-plan." >&2
+  if [ "${AGENT_ROLE:-}" = planner ]; then
+    echo "  The plan is not approved. Supervisor: read the reason ($AGENTIC_STATE/stuck/$TASK_ID.$ATTEMPT), then" >&2
+    echo "  re-dispatch the planner with a Feedback block that addresses it (fix the plan defect it names, or" >&2
+    echo "  say which changes already shipped), or fix the artifact out of band and run gate-stage2-complete." >&2
+  else
+    echo "  This attempt is not a pass. Supervisor: do not merge; discard the worktree and re-dispatch" >&2
+    echo "  $TASK_ID with a Feedback block that addresses the reason ($AGENTIC_STATE/stuck/$TASK_ID.$ATTEMPT), or re-plan." >&2
+  fi
   return 0
 }
 

diff --git a/plugin/bin/gate-stage2-complete b/plugin/bin/gate-stage2-complete
@@ -5,10 +5,14 @@
 set -uo pipefail
 BIN_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"; export PATH="$BIN_DIR:$PATH"  # md-db shim -> real binary
 GATE_NAME="gate-stage2-complete"
+# A direct call (the supervisor's own check, bin/selfcheck) is logged as a self-check; only a
+# host SubagentStop payload makes a verdict. Decided before _gatelib.sh reads the payload.
+[ -t 0 ] && export AGENTIC_SELFCHECK=1
 # shellcheck source=/dev/null
-. "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/_failclosed.sh"   # a crash blocks, never allows
+. "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/_gatelib.sh"   # fail-closed + verdict log + stuck release
 HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 : "${CLAUDE_PLUGIN_ROOT:=$(cd "$HERE/.." && pwd)}"
+[ -n "$GATE_PAYLOAD" ] || export AGENTIC_SELFCHECK=1
 
 sprint_dir="${1:-${SPRINT_DIR:-}}"
 if [ -z "$sprint_dir" ]; then
@@ -17,11 +21,19 @@ fi
 if [ -z "$sprint_dir" ] || [ ! -d "$sprint_dir" ]; then
   echo "WARN[$GATE_NAME]: no sprint dir found; stage-2 check skipped (weaker)" >&2; exit 0
 fi
+# The planner has no task contract: its stuck key is the sprint it plans and the host's agent
+# id (live e2e run 19: one planner was blocked 9 times on one unchanged reason, uncounted).
+# A re-dispatch is a new agent id, so it starts the count over.
+if [ -z "${TASK_ID:-}" ] && [ -n "$GATE_PAYLOAD" ]; then
+  _aid="$(printf '%s' "$GATE_PAYLOAD" | sed -nE 's/.*"agent_id"[[:space:]]*:[[:space:]]*"([^"]+)".*/\1/p' | head -1)"
+  if [ -n "$_aid" ]; then
+    export TASK_ID="plan-$(basename "$sprint_dir")" ATTEMPT="${ATTEMPT:-$(printf '%s' "$_aid" | tr -c 'A-Za-z0-9_-' '_')}"
+  fi
+fi
 
 # 1) No TBW placeholders anywhere.
 if grep -rl 'TBW' "$sprint_dir" >/dev/null 2>&1; then
-  echo "BLOCK[$GATE_NAME]: TBW remains in $sprint_dir (partial Stage-2 forbidden)" >&2
-  exit 2
+  fail "TBW remains in $sprint_dir (partial Stage-2 forbidden)"
 fi
 
 # 2) Every SN-NN story in stories.md has a dir with tasks.md + validate.md + plan.md.
@@ -33,13 +45,12 @@ if [ -f "$stories" ]; then
     lc="$(echo "$id" | tr 'A-Z' 'a-z')"          # s1-01
     dir="$(find "$sprint_dir" -maxdepth 1 -type d -name "${lc}-*" | head -1)"
     if [ -z "$dir" ]; then
-      echo "BLOCK[$GATE_NAME]: no story dir for $id: expected $sprint_dir/${lc}-<slug>/ (lowercase, e.g. ${lc}-count-words/) holding tasks.md, validate.md and plan.md" >&2
       other="$(find "$sprint_dir" -maxdepth 1 -type d -iname "${lc}*" | head -1)"
-      [ -n "$other" ] && echo "  found $other: rename it, e.g. git mv '$other' '$sprint_dir/${lc}-<slug>'" >&2
-      exit 2
+      fail "no story dir for $id: expected $sprint_dir/${lc}-<slug>/ (lowercase, e.g. ${lc}-count-words/) holding tasks.md, validate.md and plan.md${other:+
+  found $other: rename it, e.g. git mv '$other' '$sprint_dir/${lc}-<slug>'}"
     fi
     for f in tasks.md validate.md plan.md; do
-      [ -f "$dir/$f" ] || { echo "BLOCK[$GATE_NAME]: $id missing $f" >&2; exit 2; }
+      [ -f "$dir/$f" ] || fail "$id missing $f"
     done
   done
 else
@@ -56,32 +67,26 @@ if [ -n "$this_n" ]; then
             -exec grep -hoE '\bS[0-9]+-[0-9]+\b' {} + 2>/dev/null \
             | sed -E 's/^S([0-9]+)-.*/\1/' | sort -un | awk -v n="$this_n" '$1>n' | head -1)"
   if [ -n "$later" ]; then
-    echo "BLOCK[$GATE_NAME]: sprint $this_n references story IDs from later sprint $later —" \
-         "a sprint must be self-contained (context is cleared between sprints). Cut the" \
-         "dependency or move the story into this sprint." >&2
-    exit 2
+    fail "sprint $this_n references story IDs from later sprint $later — a sprint must be self-contained (context is cleared between sprints). Cut the dependency or move the story into this sprint."
   fi
   spill="$(find "$sprint_dir" -name '*.md' ! -name stories.md ! -name init.md ! -name output.md \
              -exec grep -lEi '(continue|continued|finish|finished|complete|completed|defer|deferred|carry|carried|spill|pick(ed)? up)[a-z]* (it |this |the rest )?(in|into|to|over to|during) (the )?(next|later|following|future|subsequent) sprint|sprint ?(n ?\+ ?1|'"$((this_n+1))"')\b' {} + 2>/dev/null | head -1)"
   if [ -n "$spill" ]; then
-    echo "BLOCK[$GATE_NAME]: $spill defers work into a later sprint — every task must" \
-         "finish inside sprint $this_n. Cut the story smaller; list the remainder under" \
-         "stories.md 'Out of scope'." >&2
-    exit 2
+    fail "$spill defers work into a later sprint — every task must finish inside sprint $this_n. Cut the story smaller; list the remainder under stories.md 'Out of scope'."
   fi
 fi
 
 # 4) Identity + referential integrity (ROADMAP §4.5): issue canonical ids for every story
 #    and task alias, check the structured references between artifacts, freeze the map.
 "$BIN_DIR/ids" freeze "$sprint_dir" \
-  || { echo "BLOCK[$GATE_NAME]: broken cross-artifact reference in $sprint_dir (see above)" >&2; exit 2; }
+  || fail "broken cross-artifact reference in $sprint_dir (see above)"
 
 # 5) Semantic change targets (ROADMAP §4.6), where the planner wrote them: each target's
 #    task, operation, path and symbol must hold against tasks.md and the real tree.
 for t in "$sprint_dir"/s*-*-*/targets.json; do
   [ -f "$t" ] || continue
   "$BIN_DIR/targets" check "$(dirname "$t")" \
-    || { echo "BLOCK[$GATE_NAME]: invalid change targets in $t (see above)" >&2; exit 2; }
+    || fail "invalid change targets in $t (see above)"
 done
 
 # 6) Structural validation via md-db (fallback to frontmatter check).
@@ -91,7 +96,7 @@ if "$BIN_DIR/ensure-tools" --resolve md-db >/dev/null 2>&1; then
   find "$sprint_dir" -type f \( -name stories.md -o -name tasks.md -o -name validate.md \
        -o -name plan.md -o -name plan-ready.md \) \
     | md-db validate --stdin-list --schema "${CLAUDE_PLUGIN_ROOT}/schemas/planning-artifacts.kdl" \
-    || { echo "BLOCK[$GATE_NAME]: md-db rejected $sprint_dir" >&2; exit 2; }
+    || fail "md-db rejected $sprint_dir"
 else
   echo "WARN[$GATE_NAME]: md-db absent; structural validation limited to dir/frontmatter checks" >&2
 fi

diff --git a/plugin/bin/targets b/plugin/bin/targets
@@ -245,7 +245,10 @@ def problems(sdir, targets, root, doc=None):
                 p.append("%s: create needs planned_symbol" % where)
             if t.get("symbol"):
                 p.append("%s: create names a planned_symbol, not an existing symbol" % where)
-            if exists:
+            # A create whose file now exists AND defines its planned_symbol is realized (its task
+            # merged; a mid-sprint re-plan re-checks the whole plan), not a defect.
+            if exists and (not t.get("planned_symbol")
+                           or locate(root, path, t["planned_symbol"])[0] is None):
                 p.append("%s: create targets %s, which already exists (use modify/extend)" % (where, path))
         elif not exists and path not in planned:
             p.append("%s: %s targets %s, which is neither in the tree nor planned" % (where, op, path))
@@ -259,6 +262,11 @@ def problems(sdir, targets, root, doc=None):
 
 
 def resolve(t, root):
+    if t.get("operation") == "create" and t.get("planned_symbol") and os.path.exists(os.path.join(root, t["path"])):
+        line, be = locate(root, t["path"], t["planned_symbol"])
+        if line is not None:                     # realized create: the planned file and symbol exist now
+            return "%s %s %s:%s %s [%s] — %s" % (t["change_ref"], t["operation"], t["path"], line,
+                                                 t["planned_symbol"], be, t.get("reason", ""))
     if t.get("operation") == "create" or not os.path.exists(os.path.join(root, t["path"])):
         return "%s %s %s %s (planned, shadow) — %s" % (t["change_ref"], t["operation"], t["path"],
                                                         t.get("planned_symbol", ""), t.get("reason", ""))

diff --git a/plugin/pipeline/planning/02-planner/persona.md b/plugin/pipeline/planning/02-planner/persona.md
@@ -49,6 +49,14 @@ hard; 150k / 20 min soft — see `docs/agents/defaults.md`). A worker still runn
 the hard limit is STOPPED by the hook and the task comes back to you to split — never retried.
 So size each task so one worker can do its RED or its GREEN well inside the soft
 limit: a handful of tests, one or two files, one behaviour. When in doubt, split.
+A later task's RED tests must fail against what the EARLIER tasks leave on main, not only
+against the empty tree: `gate-red-verify` requires every new test to fail by assertion, so
+a task whose planned tests already pass against the previous task's skeleton (a placeholder
+that maps every non-success to one error variant, a default where a `todo!()` was) cannot be
+RED and costs a mid-sprint re-plan (live run 19: three of T2's tests passed against T1's
+placeholder). Plan each later task's tests around behaviour the earlier task's contract does
+not provide, and state in the earlier task's summary what its skeleton returns for inputs
+outside its scope, so the later tests can assert something it gets wrong.
 If defaults.md sets `AGENTIC_TIME_MODE=auto`, propose ONE stable task duration for the
 project (with a min and a max) to the supervisor, who records it with `timebox propose`;
 ask for a revision only with evidence (a stats.md, an event id).

diff --git a/plugin/skills/aloop/SKILL.md b/plugin/skills/aloop/SKILL.md
@@ -397,7 +397,7 @@ in a legacy project — the gates read either):
   .aladdin/task.env        optional copy of the contract (Codex/OpenCode, older sprints);
                            it must match the contract when both exist
   .aladdin/budget-exceeded marker the time-box hook writes at a HARD limit
-main tree: .aladdin/state/stuck/<task>.<attempt>  a gate released a worker that repeated
+main tree: .aladdin/state/stuck/<task>.<attempt>  a gate released a worker (or, as plan-sprintN.<agent>, a planner) that repeated
                            one block 3 times (unverified; `budget watch` reports it once)
                            STORY_DIR is the ABSOLUTE path to sN-NN-<slug>/ (the comms dir).
   .aladdin/scaffold-symbols  scaffolder-written production-symbol list (scaffold gate)