agentleFS
davidscotttuftsdavidscotttufts/hyvemind/skills/ship-change/SKILL.md

ship-change

Skill0 starsChanged 19 days agoApache-2.0View on GitHub

Take a feature or change from idea to an open PR as one gated pipeline: spec → design review → plan → implement → review → PR, stopping at every gate. Each phase exits on a mechanical check, not on the agent's say-so. Use for any non-trivial change; start mid-pipeline by naming a phase ('ship from review').

5 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before anything else, classify the change and say which class you picked:

    line 18
  2. Distinctive:

    When unsure, go one class up.

    line 26
  3. Distinctive:

    Write the test first and watch it fail for the right reason.

    line 43
  4. Distinctive:

    When a task is large, give it to a fresh subagent with only the spec, the task and the files it needs, then review its result against the …

    line 44
  5. Distinctive:

    Keep a short running log in the PR description draft: phase, artifact, exit evidence.

    line 65
  6. Distinctive:

    Keep the spec's gap and risk registers current.

    line 66
  7. Distinctive:

    Don't try a third variation without new information.

    line 76

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “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

Ship a change

{{skill:write-spec}} → {{skill:design-review}} → {{skill:plan-tasks}} → implement → {{skill:review-change}} → {{skill:open-pr}}

This skill orchestrates the other bee skills. It reimplements none of them. Pass each phase's artifact into the next. Stop at every ⛔ gate and wait for the user. {{skill:evidence-rules}} applies throughout.

Size the process to the change

Before anything else, classify the change and say which class you picked:

ClassExamplesPhases
Trivialtypo, copy, a one-line fix with an obvious testimplement → review → PR
Containeda bug fix or small feature inside one moduleplan → implement → review → PR
Structurala new module or seam, cross-cutting behaviour, data/auth/API changethe full pipeline

When unsure, go one class up. A change that touches persistence, auth, a public API or a wire format is Structural, however small the diff.

Phases

  1. Spec with {{skill:write-spec}}. Produces a spec with testable acceptance criteria and, for Structural changes, a decision record in the project's specs.dir. ⛔ The user approves the spec.
  2. Design review with {{skill:design-review}}. Fold its corrections back into the spec. ⛔ Stop on any boundary collision, duplicated system, or unsafe data or auth path. ⛔ Build nothing until the spec (or contract) is Accepted.
  3. Plan with {{skill:plan-tasks}}. Phased tasks, each sized for one fresh-context session and each with its own check.
  4. Implement, one plan task at a time. {{cli}} run plan.md --spec <spec> automates this step: a fresh agent per task, the task's own exit check, and an independent reviewer. Or do it by hand:
    • Work in your own branch or worktree ({{skill:parallel-work}}).
    • Write the test first and watch it fail for the right reason.
    • When a task is large, give it to a fresh subagent with only the spec, the task and the files it needs, then review its result against the spec yourself.
    • Inner loop: {{cli}} affected --run runs only the tests your change can reach, and everything when it can't tell.
    • Exit check per task: the task's own check passes. Never move on with a red task.
  5. Review with {{skill:review-change}}. It starts with {{cli}} gate (full). ⛔ Stop on any blocker, go back to step 4, then review again.
  6. PR with {{skill:open-pr}}. Pushing your branch needs no approval. ⛔ The user approves the PR and any comment before they are posted.
  7. Merge. Approved PRs can share one gate with {{cli}} train <pr> <pr>…: it merges on green only if the base did not move, verifies the base's tree is the gated tree, and bisects a red train. If the project merges some other way (train.via is external), {{cli}} train refuses and prints where merges go; follow that, and never merge by hand around it.

Running it

  • Announce each phase ("Phase 3/6: plan") and show what it produced.
  • Resume anywhere. "I already have the spec" means start at design review.
  • Re-sync with the base branch at each milestone. In a shared repo, assume other sessions are pushing.
  • Keep a short running log in the PR description draft: phase, artifact, exit evidence.
  • Keep the spec's gap and risk registers current. A gap with no path to closing is a risk.
  • Never cut scope because the work is big. Split it into phases; only the user drops scope.

Stop conditions

  • A ⛔ gate trips → stop, summarize, wait.
  • A check fails → fix it before advancing. Never review or open a PR on red code.
  • Two attempts at the same failure have not worked → stop and report what you tried. Don't try a third variation without new information.

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.