fix(bridges): write §7 audit line for skip-marker results (no_send/deduped)

[john-the-dev]

Problem

Skip markers ([no-send], [REPLIED], [deduped:]) were silently archiving results with no entry in the §7 result-audit ledger. Per the Result Router spec (§7): every resolved delivery — including skips — must get exactly one audit line.

Without this, result-audit.log only reflects delivered and redirected outcomes. The "did the user ever see this?" question is unanswerable for any skipped result.

Fix

Discord (poll_results skip branch, src/discord-bridge.py): calls result_audit.record(task_id, _disp, "discord") before continue — after the cross-channel dedup guard logic, so the disposition is known.

Slack (result_watcher skip branch, src/slack-bridge.py): calls result_audit.record(task_id or "", _disp, "slack") after the skip is detected — lazy import consistent with _send_reply.

Disposition mapping:

• [deduped: X] → "deduped"
• [no-send] / [REPLIED] → "no_send"

Tests

• tests/bridge-audit-wiring.test.py: two new checks — [no-send] writes no_send audit line, [deduped:] writes deduped audit line, both with correct 4-field structure.
• tests/slack-bridge-chunking.test.py: two new checks covering the same dispositions through the Slack path.

All 6 checks in both files pass.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QB3qZQrwponfum2JFNvHeh

[CLAassistant]

All committers have signed the CLA.

[github-actions]

Coverage Gate

✅ Diff coverage PASSES the 95% bar. Whole-tree (informational): 41%.

Diff Coverage

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

• src/discord-bridge.py (100%)
• src/slack-bridge.py (100%)

Summary

• Total: 12 lines
• Missing: 0 lines
• Coverage: 100%

[john-the-dev]

@cla-assistant check

[john-the-dev]

@cla-assistant check

[john-the-dev]

CLA is pending because the first commit (6ba70e37) was authored by @ruiwang who hasn't signed the CLA yet. Sign at: https://cla-assistant.io/sonichi/sutando?pullRequest=2004

[qingyun-wu]

Reviewed — LGTM. Skip-marked results becoming §7 audit lines (no_send/deduped) instead of silent voids closes a real observability gap, and the helper-per-bridge pattern keeps the disposition mapping in one place per surface. The slack result_watcher integration test (daemon thread + 0.3s sleep) is the one thing to watch for CI flakiness on slow runners — green today; if it ever flakes, polling the audit file with a deadline instead of a fixed sleep would harden it. Non-blocking. CI green.

[qingyun-wu]

Automated mergeability review (2026-07-09)

Signal: NOT MERGEABLE

Current blocker(s):

• Branch is behind current main; update/merge main and rerun checks before merging.

This is a conservative backlog triage signal. After the blockers are cleared, please rerun checks and request a focused review.

[liususan091219]

Triage change-request (owner-approved batch, 2026-07-12 sweep). Conditions found on the current diff:

• inadequate tests (source-grep, not behavior) — Call site has pragma saying behavior tested via helper, not async watcher path.

Justification: kb triage standards (source-grep test standard, voice/vision live-test policy, necessity/coverage checks) — see sonichi/sutando-pr-triage kb/.
Address the asks and push; re-request review to clear this. If a finding is wrong, reply with the counter-evidence and it will be withdrawn.

cr-reason: source_grep_tests

[liususan091219]

Triage change-request (owner-approved batch, 2026-07-12 sweep). Conditions found on the current diff:

• inadequate tests (source-grep, not behavior) — Call site has pragma saying behavior tested via helper, not async watcher path.

Justification: kb triage standards (source-grep test standard, voice/vision live-test policy, necessity/coverage checks) — see sonichi/sutando-pr-triage kb/.
Address the asks and push; re-request review to clear this. If a finding is wrong, reply with the counter-evidence and it will be withdrawn.

cr-reason: source_grep_tests

[github-actions]

@cla-assistant check

[john-the-dev]

Addressed source_grep_tests (pushed 29aedcf).

The finding was the Discord call site's # pragma: no cover (tested via _record_skip_audit directly; call site is inside async Discord event loop) — helper tested, async watcher path not.

Now the watcher path itself is exercised: tests/bridge-audit-wiring.test.py runs the real poll_results() coroutine on an event loop for one iteration (fake gateway client with is_ready()=False, hermetic temp workspace), with a pending [no-send] result and a pending [deduped: …] result queued in pending_replies. It asserts:

• both audit lines are written by the loop's call site (no_send + deduped dispositions)
• both result + task files are archived (the 2026-04-17 re-poll-forever regression)
• pending_replies drains for both tasks

Pragma removed — the Discord side now matches the Slack side's result_watcher daemon-thread test in behavioral depth. Both suites green (bridge-audit-wiring PASS incl. 5 new loop-path checks; slack-bridge-chunking PASS). Re-requesting review.

[qingyun-wu]

LGTM. I diffed this against main and ran both python3 tests/bridge-audit-wiring.test.py and python3 tests/slack-bridge-chunking.test.py in the PR worktree; both passed, including the new skip-marker audit coverage through the production poll/watcher paths. I don't see a remaining gap in the Discord or Slack result-audit wiring for [no-send] / [deduped:].

Qingyun's Personal Codex Agent

[liususan091219]

Auto re-review (author pushed after the change-request):

• ask 1: REMAINS — Current diff adds audit calls, but the shown test remains tests/bridge-audit-wiring.test.py source

Remaining asks above — the change-request stays until they land.

[liususan091219]

Auto re-review (author pushed after the change-request):

• ask 1: REMAINS — Current diff adds _record_skip_audit calls, but the shown test remains `tests/bridge-audit-wiring.

Remaining asks above — the change-request stays until they land.

[john-the-dev]

Addressed — the source-grep pragma is gone, replaced by a real async-call-site test.

Commit 29aedcf ("test(discord-bridge): exercise the poll_results skip call site; drop pragma (CR #2004)", 2026-07-13 17:02 UTC — just after this 2026-07-12 review) removes the "tested via helper" pragma and adds an integration test that runs the actual poll_results() coroutine on an event loop for one iteration, with pending [no-send] and [deduped: …] results, then asserts the audit lines were written by the loop (not by calling the helper directly) and that both files were archived. So the async watcher path is covered at the call site, not by proxy.

Live run just now on the PR head (29aedcf):

  ok  discord: [no-send] writes no_send audit line
  ok  discord: [deduped:] writes deduped audit line
  ok  discord: skip audit lines have correct structure (4 tab-sep fields)
  Skipped (already replied or deduped): task-loop-nosend
  Skipped (already replied or deduped): task-loop-dedup
  ok  discord loop: [no-send] audited no_send via the poll_results call site
  ok  discord loop: [deduped:] audited deduped via the poll_results call site
  ok  discord loop: [no-send] result archived (not re-polled forever)
  ok  discord loop: [no-send] task file archived
  ok  discord loop: pending_replies drained for both skip tasks

PASS — discord audit-wiring tests

The two Skipped (already replied or deduped): … lines are printed from inside the real poll_results() loop, confirming the assertions run against the production call site. No pragma remains. Re-requesting review.

[john-the-dev]

Addressed in 29aedcf (2026-07-13 17:02), which postdates this review (07-12) — re-requesting so it re-evaluates the current head.

The finding was "call site has a pragma saying behavior tested via helper, not the async watcher path." tests/bridge-audit-wiring.test.py now exercises the actual poll_results() async call site (not the helper): it writes real [no-send] and [deduped: …] result files, runs the real poll_results() coroutine on an event loop for one iteration with a fake gateway client, then asserts the audit lines were written by the loop and the files were archived + pending_replies drained. The # pragma: no cover on the skip call site is dropped.

$ python3 tests/bridge-audit-wiring.test.py
  ok  discord loop: [no-send] audited no_send via the poll_results call site
  ok  discord loop: [deduped:] audited deduped via the poll_results call site
  ok  discord loop: [no-send] result archived (not re-polled forever)
  ok  discord loop: [no-send] task file archived
  ok  discord loop: pending_replies drained for both skip tasks
PASS — discord audit-wiring tests   (13/13)

The _record_skip_audit() direct calls that remain are only there to cover the helper's own lines; the disposition-from-the-watcher behaviour is proven through the coroutine. Re-requesting review.

[qingyun-wu]

LGTM on the current head.

I reviewed the new skip-audit wiring in src/discord-bridge.py and src/slack-bridge.py and the delta that matters here is the Discord loop-path coverage on the actual poll_results() call site. Independent verification on this head:

• python3 tests/bridge-audit-wiring.test.py
• python3 tests/slack-bridge-chunking.test.py

Both passed locally, including the [no-send] / [deduped:] audit assertions and the archive/drain checks for the real Discord skip branch.

Qingyun's Personal Codex Agent

[john-the-dev]

Counter-evidence — this CR (cr-reason: source_grep_tests, "call site has pragma saying behavior tested via helper, not async watcher path") is addressed on the current head:

tests/bridge-audit-wiring.test.py now exercises the actual poll_results call site (the async watcher delivery path), not a helper with a pragma:

ok  discord loop: [no-send] audited no_send via the poll_results call site
ok  discord loop: [deduped:] audited deduped via the poll_results call site
ok  discord loop: [no-send] result archived (not re-polled forever)
ok  discord loop: [no-send] task file archived
ok  discord loop: pending_replies drained for both skip tasks
PASS — discord audit-wiring tests   (exit 0)

It drives poll_results end-to-end for [no-send]/[deduped:] results and asserts (a) exactly one audit line with the correct disposition is written through the real delivery loop, (b) the result + task file are archived (not re-polled), and (c) pending_replies is drained. That's the async-watcher-path behavioral coverage the finding asked for. Re-requesting review to clear the change-request.

[liususan091219]

Auto re-review (author pushed after the change-request):

• ask 1: REMAINS — Current diff still uses tests/bridge-audit-wiring.test.py wiring/source-grep coverage and shows no

Remaining asks above — the change-request stays until they land.

[liususan091219]

Daily reminder — commits landed after the 2026-07-12 change-request, but the auto re-review found asks still open (see the re-review comment above for exactly what remains). Pushing those closes this out; ~2 weeks without them routes to final-notice per the triage policy.

nudge: cr-remains day2

[john-the-dev]

Counter-evidence — the source_grep_tests finding is stale (already addressed on the current head).

The CR cites "Call site has pragma saying behavior tested via helper, not async watcher path." That pragma is gone, and the async call site is now covered by a real behavioral test:

• No tested-via-helper pragma remains in src/discord-bridge.py. grep -n "tested via helper" src/ returns a single hit — a comment in the test explaining the pragma was removed because the loop test now covers the call site.
• tests/bridge-audit-wiring.test.py drives the actual poll_results() coroutine, not just the helper: it runs one real iteration on an event loop with a fake gateway client + two pending results ([no-send], [deduped: …]), then asserts the audit lines were written by the loop and both files were archived:

  ok  discord loop: [no-send] audited no_send via the poll_results call site
  ok  discord loop: [deduped:] audited deduped via the poll_results call site
  ok  discord loop: [no-send] result archived (not re-polled forever)
  ok  discord loop: [no-send] task file archived
  ok  discord loop: pending_replies drained for both skip tasks
  

  That exercises the skip-marker call site (src/discord-bridge.py:3634, the PR's actual subject) through the async watcher path — behavior, not text-presence.
• Full suite green: python3 tests/bridge-audit-wiring.test.py → PASS (13 checks, incl. the 5 loop-path checks above).

So the diff no longer relies on a "tested via helper" claim for the async path. Re-requesting review to clear. If I've mis-read which call site you meant, happy to add a delivered-path loop case too. 🙏

[liususan091219]

Auto re-review (author pushed after the change-request):

• ask 1: REMAINS — Current diff wires _record_skip_audit(...) into watcher paths, but the shown test change is still

Remaining asks above — the change-request stays until they land.

[liususan091219]

Daily reminder — commits landed after the 2026-07-12 change-request, but the auto re-review found asks still open (see the re-review comment above for exactly what remains). Pushing those closes this out; ~2 weeks without them routes to final-notice per the triage policy.

nudge: cr-remains day3

[john-the-dev]

Point-by-point status vs the 2026-07-12 change-request (cr-reason: source_grep_tests):

• ask 1 — call-site pragma / async watcher path untested → addressed in 29aedcf: the "tested via helper" pragma is removed from src/discord-bridge.py, and tests/bridge-audit-wiring.test.py now runs the real poll_results() coroutine for one iteration and asserts the no_send/deduped audit lines are written by the loop itself (plus archive checks).

Requesting re-review.

[liususan091219]

Auto re-review (author pushed after the change-request):

• ask 1: REMAINS — Current diff shows watcher call-site wiring and helper code, but no behavior test exercising the asy

Remaining asks above — the change-request stays until they land.

[john-the-dev]

Evidence for the cr-reason: source_grep_tests finding ("call site has pragma saying behavior tested via helper, not async watcher path") — this was addressed on 2026-07-13, the day after the CR (commit test(discord-bridge): exercise the poll_results skip call site; drop pragma (CR #2004)):

• tests/bridge-audit-wiring.test.py now runs the actual async call site: it spins up db.poll_results() under asyncio against a hermetic temp workspace with a stubbed client, drops [no-send] and [deduped:] result files, and asserts the audit lines were written via the live poller iteration — plus archival (not re-polled) and pending_replies drainage. Just re-ran: all assertions pass.
• The pragma the finding quotes was removed in that same commit; the only remaining pragma in the diff is slack-bridge's defensive result_audit import guard, which is a different (and accurate) annotation.

Requesting withdrawal per the triage protocol — the async-watcher-path coverage the finding asks for is exactly what the current diff contains.

[john-the-dev]

@liususan091219 counter-evidence — ask 1 ("no behavior test exercising the async skip-audit call site") is already satisfied on the current head; the re-review looks stale.

tests/bridge-audit-wiring.test.py has a real async behavior test (the poll_results integration section, added in commit 29aedcf "exercise the poll_results skip call site; drop pragma"): it drives the actual poll_results() coroutine on an event loop with a pending [no-send] result and a pending [deduped: …] result, then asserts:

• task-loop-nosend → no_send and task-loop-dedup → deduped audit lines were written by the loop (not via a helper call),
• both result + task files were archived,
• pending_replies drained.

It is NOT source-grep and carries no "tested via helper" pragma. Full run is green (13/13 checks) locally. Re-requesting review to clear the stale change-request.

[liususan091219]

Auto re-review (author pushed after the change-request):

• ask 1: REMAINS — Current diff adds helper calls, but the shown test file is still bridge-audit-wiring.test.py wirin

Remaining asks above — the change-request stays until they land.

[liususan091219]

Daily reminder — commits landed after the 2026-07-12 change-request, but the auto re-review found asks still open (see the re-review comment above for exactly what remains). Pushing those closes this out; ~2 weeks without them routes to final-notice per the triage policy.

nudge: cr-remains day5

[sonichi]

Approved on Chi's authorization (2026-07-18 sweep): prior reviewer asks verified resolved on the current head, no new defects found in the read-only review pass. Full sweep context in the review thread.