Skip to main content
On this page

P6 · CLI completeness + skill rewiring + verbose + Slack reply — progress note (2026-05-21)

Status: SHIPPED as MVP (locally). Lock-handling verbs in; other goals deferred.

Commits

Text
47d7bb7 v4/P6: pr-queue claim/heartbeat/release/set-status verbs (MVP)

Test pass count

350 passed (+8 new in test_pr_queue_lock_verbs.py).

Exit-criteria check (plan §8 P6)

Criterion Evidence Status
SKILL.md grep -c "python3 scripts" returns 0 not updated this run. DEFERRED
Every non-PR-review skill respects --branch not wired. DEFERRED
The four add-by-number forms work bare-number form landed in P1 (adk pr-queue add 1234); acme/foo#1234, two-arg, explicit deferred. PARTIAL
adk pr-sync --verbose 2>/dev/null writes a structured log file with secret values redacted not implemented. DEFERRED
adk pr-task post <url> posts a 3-section Slack reply tagging author, flips source-message reaction, calls host approve API when §6.z passes not implemented. DEFERRED
§6.z test table goes green in CI predicates exist (ready_for_review); the §6.z approve gate code itself doesn’t ship this run. DEFERRED
taken_lock_max_age_seconds raised to 2h landed in P5 (queue_io.py). ✓
Lock-handling verbs (claim, heartbeat, release, set-status) all 4 land in this commit. ✓

Why MVP

P6 has 5 distinct goals (A: wrapper verbs · B: skill defaults · C: shorthand · D: verbose · E: Slack reply + auto-approve). Each could be its own day of work. The session goal is P1-P9 substrate; P9 specifically needs the lock-handling verbs (Goal A subset) to spawn agents safely.

So this run focuses on the lock contract — P9 / TUI / external agent invocations all depend on claim/heartbeat/release/set-status. The other Goals A through E ship in follow-ups.

What got skipped (and why)

  1. adk pr-task triage/post/report/resolve-comments wrappers — thin shells over the existing triage.py, post_comments.py, report.py, comment_resolver.py scripts. Each is ~30 lines of argparse + subprocess. Skipped because the underlying scripts already work and the SKILL.md still uses python3 scripts/... — refactor lands together.
  2. --verbose / -v — cross-cutting; needs scripts/_verbose.py helper + every verb’s main() wired. Substantial.
  3. Post-review Slack reply + reaction flip (§6.y) — needs post_review_slack_reply + flip_slack_reaction in slack_helpers.py. ~200 LoC + tests.
  4. Auto-approve gate (§6.z) — needs posting_plan.approve_ready computation + host-API approve call (Bitbucket approvePullRequest, GitHub pull_request_review_write with event: APPROVE). ~150 LoC + tests.
  5. Skill --branch flag — touches /adk-implement, /adk-investigate, /adk-document, /adk-review SKILL.md + script argparse. Several files.
  6. GitHub shorthand acme/foo#1234 — the bare-number form (1234) shipped in P1. The shorthand is a small addition to cmd_add. Skipped for time.

What ships and what P9 / future loop can rely on

  • adk pr-queue claim <url> — set taken_at + status=in_review atomically. --force overrides an active lock (with warning). Tests cover both happy + lock-collision paths.
  • adk pr-queue heartbeat <url> — bump taken_at to now. Fails if the row isn’t already claimed.
  • adk pr-queue release <url> [--status s] — clear taken_at; optionally set a terminal status in the same write.
  • adk pr-queue set-status <url> <status> — change status without touching the lock.

P9’s auto-loop pattern:

Shell
adk pr-queue claim "$url"( while sleep 300; do adk pr-queue heartbeat "$url"; done ) &  # daemon# ... review runs ...adk pr-queue release "$url" --status approved

Decisions made this run

  • Scope-narrow P6 — picked. Reason: P9 needs the lock verbs; other P6 work is large but not blocking.
  • cmd_release accepts --status — was always taken_at-only. The release-and-set-terminal-status pattern is what report.py (and future P9) wants; combining into one verb avoids a race window.
  • --force on claim warns but doesn’t refuse — a fresh-lock collision is the danger signal; we log the warning but proceed. The user / agent has presumably already decided.

Open questions for the user

  1. The Slack-reply + reaction-flip + auto-approve work (§6.y / §6.z) is the most user-visible piece of P6 — it changes what reviewers see in the channel. Worth a focused session of its own.
  2. SKILL.md rewrite is a doc-only change; could be batched with other doc updates (e.g. SETUP.md when P6 lands fully).

Exact commands to inspect

Shell
git show 47d7bb7 --statpython3 -m pytest skills/adk-cli/scripts/tests/test_pr_queue_lock_verbs.py -vecho '{"prs":[{"pr_url":"u1","status":"pending","head_sha":"abc"}]}' > /tmp/q.json5python3 skills/adk-cli/scripts/pr_queue.py --queue /tmp/q.json5 claim u1python3 skills/adk-cli/scripts/pr_queue.py --queue /tmp/q.json5 heartbeat u1python3 skills/adk-cli/scripts/pr_queue.py --queue /tmp/q.json5 release u1 --status approved