set-body: fail-loud on byte-identical no-op + gate body-edit markers on an actual body commit (phantom v4 edits, #2333)

Goal

Close the phantom-body-edit channel surfaced in #2333 interpretation round 4: an analyzer posted epm:interpretation v4 claiming two body edits applied + verify_task_body PASS, while tasks/interpreting/2333/body.md remained byte-identical to the prior commit (929387386c) — the edits existed in no durable state. Mechanism (forensically established): the body file's mtime moved at marker time but content was unchanged — set-body was handed a file that did not contain the edits (the agent edited one copy and applied another), git no-op'd on identical bytes, and the marker recorded success anyway. Same family as the #2034 unconfirmed-stand-down record: a durable claim written without probing the durable state it asserts.

Fix (prescribed by the round-4 interpretation-critic)

At the verdict/marker-collection layer (candidate surfaces: scripts/task.py post-marker pre-post hook for epm:interpretation kinds, or the analyzer spec's landing-verification duty, or workflow_lint-class check in the /issue orchestrator's marker-collection step): an epm:interpretation vN (or any marker kind that CLAIMS body edits) must be preceded by a new set-body commit touching the task's body.md — marker-time > last-body-commit-time with a non-empty body diff — OR carry an explicit no-op declaration ("no body change this round"). Cheapest robust form: set-body itself should FAIL LOUD (or print a WARNING the caller must acknowledge) when the incoming file is byte-identical to the current body — the write is then a provable no-op and the agent learns immediately instead of at the next critic round. Also worth adding to .claude/agents/analyzer.md: the landing-verification duty (re-read the live body + confirm the new commit sha) after every set-body.

Incident

2026-08-18, task #2333 interpretation micro-round 4 (marker 08:09:26Z; caught by the round-4 Claude critic 08:14:50Z via hash-object equality against the v3 commit; corroborated by the Codex composer's mtime forensics). Cost: one wasted ensemble round + a correction round.

Candidate metadata

• target_file: scripts/task.py (set-body byte-identical no-op warning) + .claude/agents/analyzer.md (landing-verification duty)
• fingerprint: set-body-byte-identical-noop-plus-marker-body-commit-gate
• confidence: high (forensically reproduced; exact mechanism established)