AGENT SKILL

send

Make a PR, wait for CI with bounded transport retries, run the repo's required reviews, and merge only when all gates pass.

License 许可
MIT
所属仓库
yc-software/qm
最近更新
2026-11-15
Stars
★ 1.6w
Forks
2.0k

01 前言 Front Matter

name
send
description
Make a PR, wait for CI with bounded transport retries, run the repo's required reviews, and merge only when all gates pass.
requiredCapabilities
[ "egress:github.com", "egress:api.github.com", "egress:gitlab.com" ]

03 SKILL.md 正文

Babysit

Make a pull request for the change we just discussed, file it, babysit CI, fix any issues that come up, and merge once CI is fully green.

Steps

  1. Make the PR. Commit the discussed change on a feature branch (create one off the default branch if currently on it). Push it.

  2. File it. Open a pull request with a clear title and description covering why and what changed. Use gh for GitHub or glab for GitLab. Detect from the remote.

  3. Wait for CI in one blocking process, with bounded transport retries. Load the github-gitlab skill; it reports the directory holding its scripts. For GitHub, capture the PR's full head SHA and use its bundled helper, not bare gh pr checks --watch:

    node <github-gitlab dir>/scripts/watch-ci.mjs --repo OWNER/REPO --pr NUMBER --head REVIEWED_HEAD_SHA
    

    Run this with execute, an explicit sandbox_id, background:true, and authorized credentials; subscribe with background action watch and the returned process_id. Keep that sandbox ID throughout the checkout, checks, and push. Keep one watcher per PR; hand off if another agent already owns it. The helper polls internally, retries recognized transport/server errors with bounded backoff, and has a 30-minute overall deadline. Set the background lifetime longer than that. Do not wake the model repeatedly to report pending checks.

    Read the helper's JSON status, not just the exit code. Pending is not failure; TLS timeouts and connection resets are not logout or permission expiry. Failed checks, auth failures, rate limits, unknown errors, exhausted retries, and deadlines all remain nonzero. Keep set -e; never mask failure with a success marker or merge. A changed head invalidates successful checks and reviews: restart both on the new SHA. Empty, skipped, or cancelled checks do not count as success.

    For GitLab, use glab ci status --live; the GitHub helper does not cover it. Preserve its nonzero status and investigate failures before proceeding.

    Also wait for asynchronous reviewers (including Bugbot) to finish on the latest commit. Watcher success alone is never approval to merge.

  4. Run the reviews the repo requires. Follow the review process in the repo's AGENTS.md or CLAUDE.md on the current head.

  5. Fix every real issue found. Address each genuine finding from the repo's reviews, from CI, and from reviewers/bots: fix it, push, and go back to step 3. Re-review the delta, not the whole diff again — a fresh full review per round is how a small PR turns into an afternoon. For anything that needs human judgment, ask the user. Cap at ~3 fix rounds, then stop and report what's unresolved.

  6. Merge once green AND reviewed. Merge only when all checks pass, the required reviews are clean (no unaddressed findings), and there are no unresolved comments/findings — re-read checks and review threads, then merge directly (gh pr merge NUMBER --repo OWNER/REPO --squash --match-head-commit REVIEWED_HEAD_SHA / glab mr merge --squash). Do not use platform auto-merge — it fires the moment required checks pass, before async bots like Bugbot finish and before your review is done. On a branch that requires a merge queue, that command queues the PR instead; report it merged only once the PR shows as merged.

Notes

  • "Pull request" = PR/MR; same thing on either platform.
  • Don't merge with open Bugbot/reviewer findings on the current commit.
  • Weight human feedback over bot feedback; don't blindly accept everything.
  • Prefer rebase over merge for conflicts. Never amend commits.
  • Verify locally with the affected tests, not the whole suite. Run the tests covering what you changed, plus typecheck and lint, then push and let CI be the full gate. CI shards the suite across parallel runners; reproducing that serially on one machine can cost 4x the wall clock for the same signal. Run the full suite locally only when the change is broad enough that you can't tell which tests are affected.

Installed skill overrides

This workflow uses the github-gitlab seed skill's bundled watcher. Existing user-authored skills with the same name are not overwritten by seed installation. If an installed send or github-gitlab skill still says to use bare gh --watch, update that managed skill explicitly with its owner's approval. Do not claim an upstream merge updates already-published custom skills or running watcher scripts.

04 同仓库更多 Skills

yc-software
yc-software/qm

zoom

Find, create, reschedule, and cancel the user's Zoom meetings through their own Zoom connection.

★ 1.6w
yc-software
yc-software/qm

use-shared-credential

When you need to call a service the org has a SHARED credential for (a SERP/search key, a paid-API key, a data-vendor feed, a Git remote) — and the platform has told you that credential is available to this conversation — make the call BY PROXY through the credential broker. You never see the secret; the core stamps it onto the outbound request for you.

★ 1.6w
yc-software
yc-software/qm

ui-state

See what the person is looking at in the QM web UI right now — the whole page (sidebar, every multiview pane, open dialogs) as raw HTML, the app's state as JSON, optionally the CSS and a screenshot. Use when they ask about something on their screen, want help with the UI, or you need to check what a canvas you wrote actually rendered.

★ 1.6w
yc-software
yc-software/qm

ui-canvas

Draw your own HTML/CSS/JS as a full-window overlay on top of the QM web UI — a dashboard, a form, a visualization, controls that message you back. Write, update, replace, pin or dismiss it. Use when an interactive or visual result beats a chat reply.

★ 1.6w
yc-software
yc-software/qm

taste-skill

Design taste and process for anything a person will look at in a browser — landing page, dashboard, prototype, deck. The default design driver for a browsable artifact, with an anti-slop playbook for the visual language.

★ 1.6w
yc-software
yc-software/qm

slack-drafts

Create Slack message drafts in the user's own composer, as the user, through their per-user Slack OAuth. Drafting only — the user reviews and sends from Slack.

★ 1.6w