Four checker misfires from one session get fixtures and fixes

engine/hooks/gate-blame-needs-evidence/detect.py

@@ -337,8 +337,10 @@ def blamed_gates(message: str, lines: list[dict], hooks_dir: str = HOOKS_DIR) ->
     out: dict[str, dict] = {}
     for blame in blames:
         gates = gates_in(blame["window"], known)
-        if not gates and (blame["specific"] or GATE_NOUN_RE.search(blame["window"])):
+        if not gates and blame["specific"]:
             gates = latest_refusal_gate(lines) + in_reply
+        elif not gates and GATE_NOUN_RE.search(blame["window"]):
+            gates = latest_refusal_gate(lines)
         for gate in gates:
             out.setdefault(gate["name"], {"gate": gate, "phrase": blame["phrase"]})
     for delete in deletes:

engine/hooks/gate-blame-needs-evidence/tests/fixtures/real_session.json

@@ -161,6 +161,57 @@
     "label": "1527",
     "reply": "Found it. `scripts/review-unit-rules.mjs:78-85` — the `activation-surface` unit matches the bare words **`cli`** and **`command`**. `:96-100` — `proof` matches **`test`**.\n\nSo a plan that adds a CLI command can't describe itself without tripping `activation-surface`. Every rewrite I tried was me avoiding the word \"command\" in a plan about adding a command. That makes the plan worse, not more reviewable.\n\nThe design is unchanged and still right: **`invoker-cli run-chain` compiled into the binary**, skill documents the command and carries no script, watchers call the binary rather than a repo path.\n\nTwo ways forward, and I'd take the first:\n\n1. **Implement `run-chain` directly** in `packages/cli/src/index.ts` with its test, as one reviewed PR. One package, one review slice — the routing rule says that stays local, no Invoker plan needed.\n2. **Separately fix the detector**, since matching bare `cli`/`command` leaves it unable to gate any CLI work.\n\nNothing is lost meanwhile: all 11 workflows are live, and `submit-workflow-chain.sh` is untouched.\n\nGo implement `run-chain` now?"
   },
+  "silent_after_agent_routing_guard_read": {
+    "label": "agent-routing-real-generic-blame",
+    "reply": "Its source at engine/hooks/agent-routing-guard/detect.py:66 counts any bare merge word as a publication verb, and my read-only prompt said PRs 288, 289 (merged). I'll note the guard's false positive on a read-only prompt as a backlog item.",
+    "transcript": [
+      {
+        "type": "user",
+        "message": {
+          "role": "user",
+          "content": "PreToolUse:Bash hook error: [python3 $HOME/.claude/hooks/agent-routing-guard/claude_pretool_agent_routing.py]: agent-routing-guard refused a read-only prompt that mentioned merged PRs."
+        }
+      },
+      {
+        "type": "assistant",
+        "message": {
+          "role": "assistant",
+          "content": [
+            {
+              "type": "tool_use",
+              "id": "t-grep-agent-routing",
+              "name": "Bash",
+              "input": {
+                "command": "grep -n \"bare merge\" engine/hooks/agent-routing-guard/detect.py",
+                "description": "Read the agent-routing-guard detector"
+              }
+            }
+          ]
+        }
+      },
+      {
+        "type": "user",
+        "message": {
+          "role": "user",
+          "content": [
+            {
+              "type": "tool_result",
+              "tool_use_id": "t-grep-agent-routing",
+              "is_error": false,
+              "content": "66: PUBLICATION_RE = re.compile(r\"\\\\b(?:merge|merged|ship|land)\\\\b\", re.IGNORECASE)"
+            }
+          ]
+        }
+      },
+      {
+        "type": "user",
+        "message": {
+          "role": "user",
+          "content": "Stop hook feedback:\n[python3 $HOME/.claude/hooks/wrong-check-reflect/claude_stop_check.py]: wrong-check-reflect would require reflection if the assistant admitted a previous check was wrong."
+        }
+      }
+    ]
+  },
   "stop_feedback_after_refusal": [
     {
       "type": "user",

engine/hooks/gate-blame-needs-evidence/tests/test_hooks.py

@@ -176,6 +176,23 @@ def test_silent_on_real_review_unit_reply_after_its_source_was_read(self):
         reply = REAL["silent_after_review_unit_read"]["reply"]
         self.assertIsNone(detect.decide_stop_from_lines(reply, lines_of(REAL["review_unit_read"]), HOOKS_DIR))
 
+    def test_silent_on_generic_blame_after_latest_refusing_gate_was_read(self):
+        case = REAL["silent_after_agent_routing_guard_read"]
+        lines = lines_of(case["transcript"])
+        reply = (
+            "The wrong-check-reflect hook was named only in Stop feedback. "
+            "The read-only prompt evidence is below. "
+            "The guard's false positive belongs on the backlog."
+        )
+        blames = detect.blamed_gates(reply, lines, HOOKS_DIR)
+        self.assertEqual([b["gate"]["name"] for b in blames], ["agent-routing-guard"])
+        self.assertIsNone(detect.decide_stop_from_lines(reply, lines, HOOKS_DIR))
+        self.assertEqual(
+            [b["gate"]["name"] for b in detect.blamed_gates(case["reply"], lines, HOOKS_DIR)],
+            ["agent-routing-guard"],
+        )
+        self.assertIsNone(detect.decide_stop_from_lines(case["reply"], lines, HOOKS_DIR))
+
     def test_silent_when_user_ran_the_read_themselves(self):
         lines = [{"type": "user", "message": {"role": "user", "content":
                   "<bash-input>cat ~/.claude/hooks/scope-lock/detect.py</bash-input>"}}]

engine/hooks/llm-judge/phrases/wrong-check-reflect.json

@@ -16,7 +16,10 @@
     "I double-checked my earlier count and it holds; nothing in it was wrong.",
     "If my earlier check was wrong, say so.",
     "The test was wrong, not the code.",
-    "He said I was wrong about the timeout."
+    "He said I was wrong about the timeout.",
+    "I'll note the guard's false positive on a read-only prompt as a backlog item.",
+    "That refusal was a false alarm from the spawn guard, not my error.",
+    "The scan reported a false negative on the second fixture."
   ],
   "on_hit": "Wrong-check admission on this transcript. This is a FAILURE, not a preference ping: a claim went out before a real check. Finish the live correction first. Then read the reflect skill and spawn a subagent for steps 1-4 on this exact transcript. Present Accepted / Backlog / Route-to-automate-me / Rejected. Do not skip because the task also finished."
 }

engine/skills/reflect/SKILL.md

@@ -41,6 +41,10 @@ Not every finding belongs in a skill edit here. A finding about the *user's work
 
 Every invocation of this skill — single-transcript or multi-conversation mode — runs inside a subagent, no exceptions. The parent launches it with the harness subagent tool (`subagent_type: general-purpose` / `generalPurpose`: the process reads transcripts fresh from disk and doesn't need the parent's own conversation context) with the user's original reflect arguments/scope, then waits for it to report back. The subagent runs steps 1-4 (locate transcript(s), cost audit, lens fan-out, synthesis) — that's the large part. The parent runs steps 5 and 6 itself, in the main thread, never delegated for the investigation: presenting the Accepted / Backlog / Route-to-automate-me / Rejected list, then **immediately** firing a catstack worktree (+ background agent) to apply Accepted items (PR is the landing gate — do not wait for a second “apply those” turn). This keeps the bulk of the investigation out of the parent's context window — the parent only needs the final synthesized findings list.
 
+The subagent does not end its turn while a task it launched is still running;
+it waits for that task and returns once, because a resumed subagent replays its
+whole context.
+
 ## When to invoke
 
 - The `reflect-on-thrash` Stop/sessionEnd hook fired. Treat the named transcript as the scope; still present the list, then auto-worktree-apply Accepted (step 5) — never merge without a PR.
@@ -92,7 +96,9 @@ One more `Agent` call, given all reviewers' output, merges overlapping findings,
 - **Accepted** — real, durable, worth acting on. Apply the elimination hierarchy from step 3 before slotting a finding here as a skill edit: if a reviewer proposed a skill/rule fix but a categorical or lint/test fix was actually available, bump it to Backlog with the stronger fix named instead, or split it.
 - **Backlog** — real, but the right fix is higher up the hierarchy than a skill edit. Note which tier (1: categorical, 2: lint/test, 3: hook) each backlog item is.
 - **Grounding gate for skill prose.** Before an Accepted item becomes skill prose, name the established principle it instantiates — author, title, year, and a checkable URL — or write "no known prior art". An incident-shaped rule with neither goes to Backlog for grounding, not to Accepted. A rule that restates one session's bug in fresh words reads as invented and drifts into an incident log; the field's own name for it (fail fast, invariant, completeness check, reconciliation) is what the skill should say.
-- **Rejected** — one-offs, already covered, or too speculative.
+- **Rejected** — one-offs, already covered, or too speculative. A rejection
+  that says already covered by X names X at file and line; a sentence a hook
+  challenged gets evidence or a tag, never deletion.
 - **Route to `automate-me`** — real, but it's about how *this user* likes to work rather than a lesson about the code or task. Don't inline these as edits to a task-specific skill; hand the finding to `automate-me`. Same-type complaints (2+ turns or 2+ sessions) and forced iteration / product-direction change after an agent miss are **mandatory** here, not optional. Invoke `automate-me` in the same turn if the user already asked to capture the preference, or name it as the first follow-up with evidence; do not wait for them to re-prompt.
 
 **Three required parts per finding.** Every Accepted, Backlog, and Route-to-`automate-me` finding states all three. A finding missing one is not ready to sort.

engine/skills/reflect/scripts/self_retraction_scan.py

@@ -21,7 +21,7 @@
     r"read|grep|assumption|number|count)\b"
 )
 WRONGNESS_RE = re.compile(
-    r"(?i)\b(?:wrong|incorrect|inaccurate|false|untrue|not\s+true|mistaken|"
+    r"(?i)\b(?:wrong|incorrect|inaccurate|false(?![- ](?:positives?|negatives?|alarms?))|untrue|not\s+true|mistaken|"
     r"misread|mis-read|misstated|overstated|vacuous|premature|bogus|"
     r"retract(?:ing|ed)?|take\s+(?:that|it)\s+back|"
     r"does(?:n'?t|\s+not)\s+hold|did(?:n'?t|\s+not)\s+hold)\b"

engine/skills/reflect/scripts/tests/test_help_flags.py

@@ -0,0 +1,32 @@
+#!/usr/bin/env python3
+import os
+import subprocess
+import sys
+import unittest
+
+
+SCRIPTS_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+
+class TestHelpFlags(unittest.TestCase):
+    def test_token_audit_help(self):
+        proc = subprocess.run(
+            [sys.executable, os.path.join(SCRIPTS_DIR, "token_audit.py"), "--help"],
+            capture_output=True,
+            text=True,
+        )
+        self.assertEqual(proc.returncode, 0, proc.stderr)
+        self.assertIn("Usage", proc.stdout)
+
+    def test_top_sessions_help(self):
+        proc = subprocess.run(
+            [sys.executable, os.path.join(SCRIPTS_DIR, "top_sessions.py"), "--help"],
+            capture_output=True,
+            text=True,
+        )
+        self.assertEqual(proc.returncode, 0, proc.stderr)
+        self.assertIn("Usage", proc.stdout)
+
+
+if __name__ == "__main__":
+    unittest.main()

engine/skills/reflect/scripts/tests/test_self_retraction_scan.py

@@ -37,6 +37,14 @@ def test_quoted_admission_does_not_fire(self):
         text = 'The rule says "my earlier check was wrong" is an admission.'
         self.assertIsNone(self_retraction_scan.find_admission(text))
 
+    def test_false_positive_about_another_checker_stays_clean(self):
+        text = "I'll note the guard's false positive on a read-only prompt as a backlog item."
+        self.assertIsNone(self_retraction_scan.find_admission(text))
+
+    def test_false_claim_still_matches(self):
+        text = "My earlier claim was false; the real count is 12."
+        self.assertIsNotNone(self_retraction_scan.find_admission(text))
+
 
 class TestScanAssistantTexts(unittest.TestCase):
     def test_collects_one_hit_per_admission(self):

engine/skills/reflect/scripts/token_audit.py

@@ -1415,6 +1415,9 @@ def _parse_argv(argv):
     non-zero. Returns (mode, path, out_path, include_subagents)."""
     if len(argv) < 2:
         return None, None, None, True
+    if argv[1] in ("--help", "-h"):
+        print(__doc__)
+        sys.exit(0)
     mode = argv[1]
     out_path = None
     path = None

engine/skills/reflect/scripts/top_sessions.py

@@ -224,6 +224,9 @@ class of shell-portability issue.
 
 
 def main():
+    if len(sys.argv) > 1 and sys.argv[1] in ("--help", "-h"):
+        print("Usage: top_sessions.py [N]\n\nRank local sessions by total token usage.")
+        sys.exit(0)
     top_n = int(sys.argv[1]) if len(sys.argv) > 1 else 5
     t0 = time.time()
     claude_files = glob.glob(os.path.expanduser("~/.claude/projects/*/*.jsonl"))