agentleFS
aygp-draygp-dr/standard-change/.claude/skills/release-start/SKILL.md

release-start

Skill0 starsChanged 2 months agoMITView on GitHub

Say release:start on one pull request, or run the scheduler once for the oldest start. Use when a person asks to release, ship, deploy or start a change, or asks why their start has not been heard. It writes exactly one label and lets the process take over; it never deploys, approves, merges, rebases, or touches another change's markers.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read ./change/lock.sh status and gh pr list --label deploy:staging first and report the holder up front.

    line 29
  2. Very specific:

    Add deploy:staging (that asserts a deploy is running).

    line 49

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/release-start [PR | next]

The person's word. The process (change/scheduler.sh consuming through change/driver.sh) does everything after it. This skill exists to say the word correctly and to tell the person what will happen, so that a start said twice or said on the wrong change does not thrash the pipeline.

The label is release:start (renamed from change:start on 2026-09-16). Its participle, release:started, is already live and wired (the driver and scheduler write it) — confirmed 2026-09-20 against the full label list and every consuming script. What still waits for rebuild 005 is release:skip's own participle (release:skipped does not exist yet) and the secondary tier (release:hold/release:held, release:schedule).

Preconditions, checked before writing

  • The PR is OPEN and not a draft. driver.sh does not check draft; preflight's DraftGuard and unaffected.sh do, so this skill must.
  • It carries at least one app:*. With none, the word is release:skip.
  • It does not carry release:skip. unaffected.sh refuses the pair; say so.
  • It does not already carry the start label. A trigger that stays on re-fires (023 F6 started two drivers).
  • It does not carry deploy:staging: it is already in flight.
  • Read ./change/lock.sh status and gh pr list --label deploy:staging first and report the holder up front. driver exit 5 goes to stdout only (024 F3), so the person would otherwise not hear it.

The one thing it writes

gh pr edit <n> --add-label <start-label> on one PR. For next, no label: ./change/scheduler.sh --once, which takes the oldest start.

What it refuses, and says

  • "#N is a draft."
  • "#N carries release:skip; both cannot be true; withdraw one."
  • "#N has no app:* label; nothing to release; say release:skip."
  • "#N already said start at T; a second start is not a faster start."
  • "The berth is held by #M since T; your start will stay said and be retried every 20 s; nothing is asked of you."

What it never does

Add deploy:staging (that asserts a deploy is running); clear another change's markers (D21); rebase or push the person's branch (driver.sh asks the forge); approve; book a window.

How the person knows it was heard

Within two scheduler ticks: either the comment "Heard release:start. Waiting: the berth is held by #M", or the start label gone and deploy:staging on. If neither arrives in 60 s, say "no scheduler is running on this estate" (024 F4: a dead daemon leaves reassurance behind) and name the command that starts one: nohup ./change/scheduler.sh >> experiments/023-start-only/scheduler.log &.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.