daily-fix: pin #1743 acceptance grep over .claude/agents/

Overview / Motivation

Auto-filed by the /daily Step C parked-candidate sweep (2026-07-28) from a PROSE park on task #1743 (ts 2026-07-28T11:03:12Z; source: code-reviewer r1 verdict prose; no formal block, no emitter fingerprint). #1743 landed the "verdict post via --file + exact-kind read-back" change for code-reviewer + reconciler (merged 2026-07-28, commit 99af2fbb0d); its reviewer surfaced this follow-up: the acceptance criterion — no agent spec still prescribes posting a verdict body through --note argv prose — was verified by a one-shot grep during #1743 and is not pinned, so a future .claude/agents/*.md edit can silently reintroduce the argv-prose verdict-post pattern with no test failure.

Goal

Add a 1-line lint check or test pin asserting the #1743 acceptance grep stays clean (rc=1) over .claude/agents/ for the argv-prose verdict-post pattern.

Workflow gap

• Bug observed: the #1743 acceptance grep (no post-marker ... --note verdict-body prescription remaining in agent specs) ran once at review time and is unpinned; regression is silent.
• Why it is a workflow gap: .claude/agents/*.md recipes are copy-run verbatim; a reintroduced argv-prose verdict post re-opens the exact guard-block/quoting failure class #1743 closed (#1722/#1756 family).
• Confidence (emitter): low-medium
• verified-at-filing: grep -n 'argv-prose\|post-recipe' scripts/workflow_lint.py tests/test_workflow*.py → 0 hits (2026-07-29 UTC) — no such pin exists (absence claim; 0-hit in-target result IS the evidence). Landed-fix history check: git log --oneline --since='7 days ago' -- scripts/workflow_lint.py shows the #1743 merge itself (99af2fbb0d) but no pin-test commit for the acceptance grep. unverified hypothesis — verify at plan time: the exact acceptance-grep pattern #1743's session ran (recalled from the #1743 code-review round, not read from a durable artifact) — the planner should recover it from #1743's events.jsonl / merged diff before writing the pin.

Proposed change (candidate diff sketch — refine in planning)

Add either a workflow_lint.py check (bundled into the no-flags default run) or a tests/ pin test that greps .claude/agents/ for the argv-prose verdict-post pattern and FAILs on any hit, mirroring the existing lint-pin idioms (e.g. --check-piped-git-push).

Scope / surfaces

• Primary target: scripts/workflow_lint.py (alternatively a tests/test_workflow_lint*.py pin)
• Recover the acceptance grep from #1743's merged diff (99af2fbb0d) / events before implementing; grep .claude/agents/ to confirm current rc=1 baseline.

Constraints / invariants

• Workflow-surface only. scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
• New no-flags-bundled checks follow the c37 pin-test convention where applicable.
• This session runs under EPM_WORKFLOW_FIX_SESSION=1 / a workflow_fix_target: Provenance line — recursion guard applies.

Provenance

• fingerprint: d414fbf24fd1

• workflow_fix_target: scripts/workflow_lint.py

• fingerprint: (driver-computed; prose park — no emitter fp)

Verbatim surfaced prose (park note on #1743, 2026-07-28T11:03:12Z): "parked — running under workflow_fix_target recursion guard [...] Source: code-reviewer r1 verdict prose (task #1743). Candidate: add a 1-line lint/test pin asserting the acceptance grep stays clean (rc=1) over .claude/agents/ for the argv-prose post-recipe pattern; target_file: scripts/workflow_lint.py (or a tests/ pin). Confidence: low-medium; nothing filed (nightly /daily Step C sweep is the route)."