fix(bg): B-0440.3 — PR-activity poll (closes Riven's P0 AC violation; 16 tests pass)

[AceHack]

Summary

Resolves Riven's adversarial review (bus envelope `6c689634-14e7-4cf9-acf8-00c018f1bded`). Fixes both flagged findings:

• P0 (AC violation): Standing-by detector now reads BOTH commit history AND PR activity. Was previously commit-only — false negatives on agents doing PR-review-only / bus-coordination / claim-work.
• P1 (silent failure): Bus publish failure now surfaces in structured `lastPublishError` field, not just buried in note string.

Real smoke test

```
{
lastCommitAt: 2026-05-13T18:49:06.000Z,
lastPrActivityAt: 2026-05-13T19:17:58.000Z,
idleMinutes: 1.08, // gap from MAX of the two signals
publishedEnvelopeId: 606cae9e-...,
lastPublishError: null,
}
```

Substrate-honest framing

The detector is repo-level (no `--author` filter) because factory agents share the AceHack GitHub account. Author-filtering would miss most activity. Documented in adapter docstring.

Tests

```
bun test
16 pass
0 fail
47 expect() calls
```

Multi-agent coordination footnote

Riven flagged this gap in a 5-finding adversarial review and routed it via bus envelope. Otto replied on the bus (envelope `e8174b34-fdee-47f7-af1a-df80c27b51cd`) accepting all findings and pivoting to fix P0 first. This PR is the operational result of that bus coordination — no human courier in the loop.

The factory walks.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR updates the background standing-by detector (tools/bg/) to treat recent PR activity as “work” (in addition to commits) when deciding whether an agent is idle, and makes bus publish failures machine-readable in the poll result.

Changes:

• Add gh pr list-based PR activity polling and compute idle time from the most-recent of (last commit, last PR activity).
• Extend PollResult with lastPrActivityAt and structured lastPublishError surfaced on publish failures.
• Update unit tests to cover PR-only activity, mixed commit/PR cases, and publish-failure reporting.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
tools/bg/standing-by-detector.ts	Adds PR-activity signal + structured publish error field; updates idle computation and output fields.
tools/bg/standing-by-detector.test.ts	Updates tests to validate MAX(commit, PR) behavior and structured publish failure reporting.

Comments suppressed due to low confidence (1)

tools/bg/standing-by-detector.ts:136

• The pollOnce docstring introduces named attribution (“Riven’s P0 finding”). For code under tools/, the repo convention is role-refs only outside history surfaces (.github/copilot-instructions.md:305-362). Please rewrite to remove the persona name while keeping the envelope/issue reference.

 * Resolves Riven's P0 finding (envelope 6c689634-...): the prior slice-2
 * version only checked commit-history and produced false negatives on
 * non-commit agent activity.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c0ba0918cb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Query PR activity by updated timestamp, not creation order

Using gh pr list --state all --json updatedAt --limit 1 does not guarantee you get the most recently updated PR; gh pr list is ordered by PR creation time (GitHub CLI’s PullRequestList query uses orderBy: {field: CREATED_AT, direction: DESC}), so activity on older PRs can be missed. In a repo where agents review/comment on an older PR while no new PRs are created, lastPrActivityIso can return a stale updatedAt, causing false idle detection and incorrect nudges.

Useful? React with 👍 / 👎.