[sonichi]

What's wrong

src/dm-result.py is the REST-only fallback delivery path. poll_dm_fallback (discord-bridge) shells out to it for any results/task-*.txt that no channel consumer claimed within 90s — making it the last consumer in the result chain.

It already parses [file:|send:|attach:], and its own comment states why:

without parsing these markers it would deliver the literal text [file: /tmp/sutando-x.png] in the DM

It did not parse the skip markers — [no-send], [REPLIED], [deduped: …] — and that argument is stronger for them. A [no-send] body is not merely ugly in a DM; its entire purpose is that it must not be delivered. Because this script is last in the chain, a marker it ignores becomes exactly the DM the marker existed to prevent, with the literal [no-send] text included.

CLAUDE.md documents [no-send] as "Discord bridge skips delivery for this task (still archives)", and the bridge's WS path honors it. Ten modules import result_markers — both bridges, task-bridge, slack, telegram, morning-briefing, result_router, local_task_protocol, reply_chain, discord_addressee. dm-result was the one consumer in that family that didn't.

How I hit it

Not hypothetically. I was about to write a probe task for an unrelated evidence capture and checked whether its result would reach the owner. It would have — task-* is in FALLBACK_PREFIXES, and [no-send] would not have saved it.

The change

Parse markers and return early on a skip action, before the voice probe — a body marked do-not-deliver shouldn't be delivered by any path, and returning early also skips a needless SSE call.

Evidence

The new suite at this head and at the parent, tests unchanged:

at this head:
  Ran 6 tests    OK

at the parent (fix reverted):
  Ran 6 tests    FAILED (failures=3)
    test_no_send_is_not_delivered   a [no-send] body was delivered to the owner's DM
    test_replied_is_not_delivered   a [REPLIED] body was delivered again
    test_deduped_is_not_delivered   a [deduped:] body was delivered

The three controls pass in both states, so the suite isn't vacuous:

• an ordinary body is still delivered
• a [file:] body is still delivered (attachment handling unaffected)
• prose that merely mentions [no-send] mid-sentence is still delivered — the marker only counts at the start of the body, so this does not reintroduce the failure mode fix(markers): stop [dm-only] editing prose that merely mentions it #2481 fixes for [dm-only]

Each case asserts the marker path is taken without reaching the network — the test never supplies a token, so a real send attempt would be visible.

🤖 Generated with Claude Code

[github-actions]

Coverage Gate

✅ Diff coverage PASSES the 95% bar. Whole-tree (informational): 70%.

Diff Coverage

Diff: origin/main...HEAD, staged and unstaged changes

• src/dm-result.py (100%)

Summary

• Total: 5 lines
• Missing: 0 lines
• Coverage: 100%

[qingyun-wu]

Reviewed current head be25b854e84fe4c90241de5eb5a98a42258590cc.

No blocking findings in the code or local focused checks. dm-result.py now consumes the shared result-marker parser before the voice/network delivery path and exits on terminal skip actions, so [no-send], [REPLIED], and [deduped: ...] bodies cannot leak through the last-resort DM fallback when no earlier consumer claimed the result. Plain bodies, file-marker bodies, and prose that merely mentions a marker still reach the delivery path.

Checks: git diff --check origin/main...HEAD, py_compile with PYTHONPYCACHEPREFIX, python3 tests/dm-result-skip-markers.test.py, python3 tests/result-markers.test.py, python3 tests/bridge-marker-no-leak.test.py, python3 tests/bridge-audit-wiring.test.py, python3 -m ruff check src/dm-result.py tests/dm-result-skip-markers.test.py, and bash tests/review-checks.test.sh passed. GitHub checks are mostly green, with tsc + tests (clean install) still pending, so I am leaving this as a review comment rather than an approval until that current-head CI finishes.

Reviewed by Qingyun's Personal Codex.

[qingyun-wu]

Reviewed current head be25b854e84fe4c90241de5eb5a98a42258590cc.

No blocking findings. The current-head CI that was still pending during my first pass has completed green, and the local focused checks already covered the skip-marker boundary. dm-result.py now honors terminal result markers before the voice/network delivery path, so [no-send], [REPLIED], and [deduped: ...] bodies cannot leak through the last-resort DM fallback; ordinary text, file markers, and prose merely mentioning a marker remain deliverable.

Checks: git diff --check origin/main...HEAD, py_compile with PYTHONPYCACHEPREFIX, python3 tests/dm-result-skip-markers.test.py, python3 tests/result-markers.test.py, python3 tests/bridge-marker-no-leak.test.py, python3 tests/bridge-audit-wiring.test.py, python3 -m ruff check src/dm-result.py tests/dm-result-skip-markers.test.py, bash tests/review-checks.test.sh, and current-head GitHub CI passed.

Reviewed by Qingyun's Personal Codex.

[qingyun-wu]

Reviewed exact head be25b854e84fe4c90241de5eb5a98a42258590cc after the previously pending CI completed. The full dm-result.py fallback path now parses shared result markers before voice/network delivery and returns on terminal skip actions, so [no-send], [REPLIED], and [deduped: ...] cannot leak through the last-resort DM consumer. Ordinary bodies, attachment markers, and inline prose mentions still deliver.

Reviewer-run tests/dm-result-skip-markers.test.py (6/6), tests/result-markers.test.py (34/34), bridge marker/audit suites, Ruff, diff check, and REVIEW.md path scan pass. All exact-head GitHub checks and CLA are green. The worst-case regression—silently suppressing ordinary output—is bounded by the positive delivery controls.

Verdict: approved; code-ready on this head.

[john-the-dev]

APPROVE — verified independently.

What I checked

• Confirmed the API the fix leans on: result_markers.parse_markers(text).actions yields Action(kind="skip", value∈{no-send,REPLIED,deduped}), and skip markers are start-anchored (^\s*\[no-send\] etc.) and terminal (src/result_markers.py:103-107, 138-186). So next(a for a in ….actions if a.kind=="skip") correctly fires only when the body starts with a skip marker — prose that merely mentions [no-send] is untouched.
• Placement is right: the check sits after text is assembled and before voice_connected() and any send path, so the last-consumer-in-the-chain leak the PR describes can't reach the DM. Early-return matches the existing voice/file-marker early-return style in the same function.
• Ran the new test on the PR head in an isolated worktree: 6/6 pass — both directions covered (no-send/REPLIED/deduped suppressed; plain body, file-marker body, and prose-mentioning-marker body still delivered).
• CI: all green (diff-cover ≥95%, tsc+tests, smoke, CLA met).

Correct, minimal, well-tested, closes a real do-not-deliver leak. LGTM.

[github-actions]

@cla-assistant check