メニュー
Single iteration of PR tending. Reads PR state, classifies new events, routes fixes, reports status. Called by /loop via /tend-pr setup. Can be invoked manually to run a single tick.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
End-to-end autonomous execution: figure-out → define → do, chained without manual approval gates. Use when you want to define and execute without intervention during planning. Triggers: auto, autonomous, end-to-end, just build it, tend pr, babysit pr.
Manifest builder. Turns shared understanding into a verifiable Manifest with Deliverables, Acceptance Criteria, Global Invariants, and Approach. Use when planning features, scoping refactors, debugging complex issues, or whenever a task needs structured thinking before coding. Triggers: define, manifest, scope, plan, spec out, break down.
Manifest executor. Works through Deliverables verifying every Acceptance Criterion and Global Invariant. Use when executing a manifest, running a plan, implementing a defined task. Triggers: do, execute, run the manifest, implement plan, run plan, execute manifest, ship the manifest.
Completion marker for the /do workflow. Outputs a plain-prose summary of what was built. Called by /do after every Acceptance Criterion and Global Invariant verifies PASS, not directly. Triggers: completion summary, manifest complete, all criteria pass, finish manifest, summarize completion, wrap up, mark done.
Structured escalation when /do hits an unrecoverable blocker. Surfaces what was tried, why it failed, and what the user can decide. Called by /do, not directly. Triggers: blocked, can't proceed, unrecoverable failure, need user decision, blocker, hit a wall, escalate, stuck.
Figure things out together — any topic, problem, or idea. Presses relentlessly until shared understanding is reached. Use when you need to understand something before acting, or when figuring it out IS the goal. Triggers: figure out, help me think through, dig deeper, press this, what is really going on, investigate, work through, why does.
| name | tend-pr-tick |
| description | Single iteration of PR tending. Reads PR state, classifies new events, routes fixes, reports status. Called by /loop via /tend-pr setup. Can be invoked manually to run a single tick. |
| user-invocable | true |
Execute one iteration of PR tending: read current state, classify new events, route fixes, update PR, report status. Designed to be called repeatedly by /loop.
$ARGUMENTS = <pr-number> <mode> [<manifest-path> <log-path>]
<pr-number>: Required. The PR to tend.<mode>: manifest or babysit.<manifest-path> and <log-path>: Required when mode is manifest. Path to the manifest and execution log.If arguments missing or malformed: error and halt with usage message.
Use a lock file at /tmp/tend-pr-lock-{pr-number}. Skip this iteration if the lock exists and isn't stale (older than 30 minutes). Remove stale locks. Create the lock at iteration start, remove at end.
Read the PR's current state: open/closed/merged/draft, CI status, review status, new comments since last check, merge conflict status, unresolved threads.
Terminal states — Merged, Closed, or Draft. Handle per Output Protocol (end the loop).
First tick (tend-pr-log has no prior iterations): Everything is unprocessed — run the full pipeline.
Check categories:
Runs every tick.
Compare against base branch first:
Runs every tick.
Update the PR branch from the base branch. Preserve review comment history — rebase destroys it (see Gotchas). Flag ambiguous conflicts to the user.
Runs when new comments exist since last check. Skip when no new comments.
Label source first (bot vs human — read ../tend-pr/references/known-bots.md), then classify intent (read ../tend-pr/references/classification-examples.md):
Runs when there are CI failures to handle or when new comments were classified.
Manifest mode: This is the same default-to-amend reflex documented canonically in do/SKILL.md Mid-Execution Amendment — PR comments and CI failures are external feedback that contradicts or extends the manifest, so they route through Self-Amendment. The only difference is the trigger source (PR thread vs. user message in a session).
Repos: (multi-repo) and deliverables carry repo: tags, look up this PR's repo via the platform adapter (e.g., GitHub MCP returns owner/repo for the PR number); for each absolute path in the manifest's Repos: map, run git -C <path> remote get-url origin and parse the URL to its owner/repo form; the entry whose remote matches the PR's owner/repo is the corresponding repo. Use that entry's name: as the deliverable filter — prefer deliverables tagged with the matching repo name./define --amend <manifest-path> --from-do. Multi-repo: the manifest is a shared canonical /tmp file used by every repo's /tend-pr-tick — concurrent amendments are last-writer-wins (no locking; see tend-pr/SKILL.md §Multi-Repo PR Sets and define/references/MULTI_REPO.md §f). If you notice a missed amendment, re-trigger the lost tick./do <manifest-path> <log-path> --scope <affected-deliverable-ids>. If /do escalates with "Deferred-Auto Pending": implementation is green — proceed to steps 4 and 5 normally (push the fix; reply to the originating thread). Append the deferred-auto reminder to this tick's Status Report so the user sees it: "Implementation green; run /verify <manifest> <log> --deferred when prerequisites are in place to reach /done." Do NOT terminate the tick — subsequent ticks continue normally toward merge-readiness. If /do escalates with any other type: log the blocker, report the escalation to the user, and end the loop.Babysit mode: Fix directly, push. If the item originated from a comment, reply on that thread. (No manifest in scope = no amendment route; this is the no-manifest case from do/SKILL.md — fail-open inline handling.)
False positives: Reply with explanation.
Uncertain: Reply asking for clarification, leave thread open.
Thread resolution rule: Resolve all bot threads after addressing (fix, reply, or both). Never resolve human threads — the reviewer owns their thread and will resolve it themselves.
Runs when this tick produced changes (routing fixes, conflict resolution). Skip when no changes were made.
After changes, rewrite "what changed" sections to reflect the current diff. Preserve manual context (issue references, motivation, deployment notes). Update title if scope changed significantly.
The PR description stays summary-only — never embed the manifest. Manifests are internal working documents (also true in multi-repo, where the same canonical manifest is shared across all PRs without surfacing to reviewers).
Append to /tmp/tend-pr-log-{pr-number}.md: timestamp, actions taken, skipped items, remaining blockers, current PR state.
Runs every tick.
When the PR's merge state indicates it is mergeable (all required checks pass, required approvals obtained, no unresolved threads, no pending /do runs) — this is a terminal state. Ask the user about merging and end the loop per the Output Protocol.
Unresolved uncertain threads block merge-readiness — they represent unanswered questions that could surface actionable issues.
Determine merge requirements from the platform's merge state (e.g., GitHub branch protection rules), not hardcoded assumptions about what's required.
Stale thread escalation: If an uncertain thread has received no reply for 30+ minutes, or a fixed thread remains unresolved for 30+ minutes, escalate to the user: "Thread from @reviewer unresolved for [duration]: [uncertain — no reply / fixed — awaiting reviewer resolution]. Continue waiting, resolve, or ping reviewer?"
Every iteration MUST end with exactly one of these outcomes. The tick owns the full lifecycle — when a terminal state is reached, handle it completely (user interaction + loop termination) before returning.
When any STOP condition is reached: handle the user interaction described below, then end the loop — do not call ScheduleWakeup or CronCreate for the next iteration.