Autoship logo

Autoship

Community
mblode
autoship

Runs a changesets npm release through the version PR, CI publish, and registry verification. Use when asked to "release this package", "autoship", "merge Version Packages", or diagnose a release that did not publish. For feature PRs use pr-creator or pr-babysitter.

Overview

Publishermblode
Repositoryagent-skills
Skill nameautoship
Stars
118
Forks
11
Bundled files
4
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 4 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by mblode on GitHub. Read the source before you install it.

Installation

Install the Autoship AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/mblode/agent-skills.git /tmp/agent-skills
mkdir -p .claude/skills
cp -r /tmp/agent-skills/skills/autoship .claude/skills/autoship
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Autoship in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Autoship on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Autoship is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Autoship

Drive an npm release end to end: changeset, fix loop, push, CI watch, Version Packages PR merge, publish watch, npm verification.

  • IS: the full release pipeline for an existing changesets-based npm package, from writing the changeset file to confirming the new version on the registry, plus diagnosing why a release run did not version or publish.
  • IS NOT: opening a feature PR (use pr-creator), monitoring a feature PR for reviews, conflicts, or CI (use pr-babysitter), general build or type fixes outside a release flow, or scaffolding a new package (use scaffold-cli, which hands off to autoship for the first release).

The Release Loop

One workflow, two successive runs. Misreading it as two workflows causes most autoship mistakes.

  1. Push a commit containing a pending .changeset/*.md file to the default branch.
  2. The release workflow runs: changesets/action sees pending changesets, runs changeset version in CI, and opens or updates a PR on branch changeset-release/<default-branch> (title "Version Packages", suffixed "(next)" in pre mode) carrying the package.json bump and CHANGELOG.md entry.
  3. Merge that PR once every check is green.
  4. The same workflow runs again. With no pending changesets left, the action runs its publish script (changeset publish), which publishes to npm and, by default, pushes the git tag and creates a GitHub release.

The local job ends at "push the changeset file". CI owns versioning and publishing; anything versioned locally breaks the loop (see Gotchas).

The action has two live majors with different input names: @v1 takes publish:, @v2 takes publish-script:. Read the uses: line before diagnosing a run that versioned but never published.

Reference Files

FileRead when
references/changeset-and-commit.mdSteps 1-3: writing the changeset file, discovering and running gates non-interactively, staging the release commit
references/ci-polling.mdStep 4 and Step 5 watches: the Monitor tool, the commit watch script, failure classification, the Changeset Status check
references/version-pr-and-publish.mdOnce CI is green: workflow shape, finding and merging the Version Packages PR, the publish run, npm verification, publish failure diagnosis
evals/evals.jsonOnly when changing this skill; never loads during a release

Intent Map

IntentStepsNotes
Full autoship (ship / release / publish)1-5Default entry point. End to end through publish, no intermediate prompts
Create changeset only1Stage a release without pushing
Fix gates and push1-3Changeset, fixes, commit, no CI watch
Watch CI only4-5Changeset already pushed
Merge Version Packages PR only4b-5CI already green; merges once preconditions hold
Fix gates only2Inside a release flow; no changeset needed
Diagnose a release that did not publishread-onlyFailure Recovery table plus references/version-pr-and-publish.md

"Ship it" with no npm release context routes to pr-creator.

Safety Tiers

Invoking autoship is standing consent for the full release flow. Do not pause mid-flow to re-confirm; gate risky steps on objective preconditions instead.

  • Green (execute directly): reads: CI and PR state, npm view, pending changesets, package.json scripts, git history and status.
  • Yellow (announce in one line, then execute): writing changeset files, running lint/type/test/format fixers, git add/commit/push, starting watches, and merging the Version Packages PR once its identity is confirmed and every check passes.
  • Red (explicit confirmation required): force-pushes, history rewrites, repository settings changes, any destructive git operation.

Workflow

Copy this checklist to track progress:

text
Autoship progress:
- [ ] Step 1: Create changeset (default patch)
- [ ] Step 2: Fix lint, types, tests, format
- [ ] Step 3: Commit and push the changeset (never `changeset version` locally)
- [ ] Step 4a: Watch CI on the pushed commit
- [ ] Step 4b: Find and merge the Version Packages PR
- [ ] Step 5: Watch the publish run, verify on npm

Step 1: Create changeset (default patch)

  • Load references/changeset-and-commit.md.
  • Inspect pending changesets and their package coverage. Reuse those covering the requested release; add one only for uncovered changes. Ask only if unrelated pending releases make the publish scope ambiguous.
  • Default to patch; minor or major only on explicit user instruction.
  • Write the file directly (the interactive prompt needs a TTY); the summary is user-facing changelog text inferred from git log --oneline -10.
  • npx changeset status validates the file: a misspelled package name fails here instead of in CI.

Step 2: Fix lint, types, tests, format

  • Discover commands from package.json scripts (check, lint, typecheck, test, format, fix); in non-npm repos check Makefile, Cargo.toml, pyproject.toml, go.mod.
  • Run lint, typecheck, test, format. After any code change, re-run from the first gate: a type fix routinely breaks lint, and a lint autofix can break a test.
  • Scope auto-fixers to changed files where supported, then check git status: broad fix/format scripts reformat files outside the change (MDX is a frequent casualty). Undo only fixer changes introduced by this run, preserving pre-existing edits in the same files.
  • Cap the loop at 5 fix iterations per gate, reporting the remaining error count each pass; then stop and report (Failure Recovery).

Step 3: Commit and push the changeset

  • Stage the changeset file and in-scope fixes by explicit path; sweep git status --porcelain for hook artifacts (a root schema.gql is a known one) before committing.
  • Commit (chore: add <type> changeset for <package>) and push.
  • The pushed commit must still contain .changeset/*.md. Running changeset version locally consumes it (see Gotchas).

Step 4a: Watch CI on the pushed commit

  • Load references/ci-polling.md.
  • Start the commit watch on the pushed SHA. It emits a line per state change and one TERMINAL: line when every run for that commit (CI and the release workflow) completes. An idle first poll is normal; runs take time to queue.
  • On failure, classify from gh run view <id> --log-failed: flaky or infra gets gh run rerun <id> --failed (max 3); a real failure gets a fix, commit, push, and a fresh watch on the new SHA.

Step 4b: Find and merge the Version Packages PR

  • Load references/version-pr-and-publish.md.
  • Find the open PR whose head is changeset-release/<default-branch>. If absent, run the PR-wait watch (10-minute cap).
  • Merge only when all three hold:
    • Head branch is changeset-release/<default-branch> (title normally "Version Packages", but pr-title can rename it). Never merge any other PR.
    • gh pr checks <n> --json name,bucket reports bucket: pass for every check.
    • gh pr view <n> --json mergeable reports MERGEABLE (on UNKNOWN, wait and re-query).
  • Announce ("Merging Version Packages PR #N: @"), then gh pr merge <n> --squash --delete-branch, or the merge method the repo allows. Any failed precondition: stop and report, never merge.

Step 5: Watch the publish run, verify on npm

  • Merging triggers the same workflow again; with no pending changesets it publishes.
  • Take the merge SHA (gh pr view <n> --json mergeCommit --jq .mergeCommit.oid) and run the commit watch on it.
  • On failure: read the log, match it against the publish failure table in references/version-pr-and-publish.md, report, and stop. Publish failures are never retried blind; every cause on that table needs a config or settings change.
  • On success: npm view <package>@<version> version must print the merged version, and npm view <package> dist-tags must show latest pointing at it (unless pre mode). Stop remaining watches and report both outputs; they are the completion evidence.

Failure Recovery

Failure pointResponse
Gate still failing after 5 iterationsStop. Report the gate, remaining error count, last error output
CI fails after the changeset pushFlaky or infra: gh run rerun <id> --failed, max 3. Real: fix, push, fresh watch
"Changeset Status" check failsNo changeset: Step 1. Consumed (a local changeset version ran): revert the bump and CHANGELOG.md edit, re-add the changeset file. Rerunning cannot fix consumed state
Version Packages PR absent after 10 minutesgh run view the release run: "not permitted to create or approve pull requests" means the repo setting is off (Gotchas). Otherwise confirm pending changesets on the default branch and a changesets/action step in .github/workflows/
Release run green but nothing publisheduses: changesets/action@v2 with the v1 publish: input, or no publish input at all. Check the run's "Unexpected input(s)" warning
Merge precondition failsStop and report. Never override failing checks or resolve conflicts in the bot PR
Publish run failsMatch the log against the publish failure table; report the fix; stop

Gotchas

  • Never run npx changeset version locally. It consumes .changeset/*.md, so the pushed commit has no pending changeset, "Changeset Status" fails, and no Version Packages PR opens. Recovery is reverting the bump, not rerunning CI.
  • Never run npm publish directly. It bypasses changesets, skips the changelog and tag, and leaves the Version Packages PR describing an already-shipped version, which then fails with "You cannot publish over the previously published versions".
  • Never hand-edit CHANGELOG.md or the package.json version. CI generates both in the Version Packages PR; local edits make the bot PR CONFLICTING.
  • changesets/action@v2 renamed every input (publish to publish-script, version to version-script, title to pr-title, commit to commit-message, branch to pr-base-branch). The old names are ignored with only a warning, so a @v2 workflow still using publish: opens the Version PR and then completes green without publishing anything.
  • setup-node with node-version: 22 ships npm 10.9.x. Trusted publishing needs npm 11.5.1 or later, so changeset publish fails ENEEDAUTH even with id-token: write set. Use Node 24 or add npm install -g npm@latest before publishing.
  • A workflow authenticating with an NPM_TOKEN secret that holds a classic token fails ENEEDAUTH: npm revoked every classic token on 9 December 2025. Move to trusted publishing rather than minting a new token; write-capable granular tokens expire within 90 days.
  • npm does not validate a trusted publisher when you save it. A workflow filename that differs from .github/workflows/<file>.yml by case or extension surfaces only at publish time, as ENEEDAUTH.
  • New personal repos block Actions from opening PRs. The release run fails with "GitHub Actions is not permitted to create or approve pull requests" and no Version PR appears. Fix: Settings, Actions, General, "Allow GitHub Actions to create and approve pull requests" (an org-level setting can override it). Red tier: report it, do not change settings unasked.
  • gh pr list --json headBranch and gh pr checks --json conclusion are invalid fields and error. Use headRefName and bucket.
  • The Monitor tool's default timeout_ms is 300000 (5 minutes). A CI run that outlasts it kills the watch with a timeout that looks like silence. Pass timeout_ms: 3600000 for CI watches.
  • Poll every 30 seconds or slower. Faster loops burn the GitHub API rate limit (gh api rate_limit --jq .resources.core.remaining) and stall the flow mid-release.
  • git add -A commits pre-commit-hook artifacts and fixer churn into the release commit. Stage explicit paths.
  • major without explicit instruction signals breaking changes to every consumer. Default to patch.

Related Skills

  • scaffold-cli: scaffolds a new TypeScript package with changesets and OIDC publishing, then hands off to autoship for its first release.
  • pr-creator: opens feature PRs. Autoship merges only the bot-opened Version Packages PR.
  • pr-babysitter: watches feature PRs (reviews, conflicts, CI). Autoship watches release CI only; never babysit a Version Packages PR autoship is driving.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Autoship AI skill do?

Runs a changesets npm release through the version PR, CI publish, and registry verification. Use when asked to "release this package", "autoship", "merge Version Packages", or diagnose a release that did not publish. For feature PRs use pr-creator or pr-babysitter.

Why use Autoship on TypingMind?

Because you install it once and use it with any model. Autoship is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Autoship in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/mblode/agent-skills/tree/main/skills/autoship. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Autoship?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Autoship?

As many as you like. As long as a model supports skills, you can use Autoship with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Autoship AI skill free?

Yes. It is published on GitHub by mblode under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇