agentleFS
neurawork-gitneurawork-git/n8n-autopilot/skills/plugin-testing/SKILL.md

plugin-testing

Skill18 starsChanged 42 days agoMITView on GitHub

The ONLY supported way to test changes to the n8n-autopilot plugin itself: commit → push to the private repo → install FROM that repo → restart → verify registration. Use whenever a plugin change (agent, skill, hook, script) must be exercised in a real session. Hard rule: no cache hand-copying, no directory-pointer marketplace, no hand-edited settings.json. Push to repo, install from there, nothing else.

  • Commits and pushes

3 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

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 3

    Distinctive

    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:

    Never add a directory source for this name (that is the cache-drift trap).

    line 58
  2. Distinctive:

    Run the verification probe — a short Workflow that confirms the install is real:

    line 72
  3. Distinctive:

    Every agent resolves (no "agent type not found") ⇒ registration clean.

    line 83
  4. Distinctive:

    If dual/stale installs or a directory-pointer marketplace exist:

    line 88
  5. Very specific:

    Never resolve a messy state by editing cache files or settings.json by hand — only the CLI (claude plugin disable/enable/uninstall, claude …

    line 97

“Common” is measured: kept by at least 6 in 10 of the 3 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

Plugin Testing — push to repo, install from there, nothing else

THE RULE. A change to this plugin (agent def, skill, hook, script, manifest) is tested in exactly one way: commit it, push it to neurawork-git/n8n-autopilot-internal, install the plugin FROM that GitHub repo, restart, verify. Nothing else counts. A change you have not pushed-and-installed has not been tested — you are looking at stale or inconsistently-registered state.

Why (lessons paid for)

  • Hand-copying files into ~/.claude/plugins/cache/.../<version>/ registers inconsistently: agents silently fail to appear, /reload-plugins does not pick up newly-added agents, and you end up with two plugins thinking different things. NEVER do it.
  • A directory-pointer marketplace (extraKnownMarketplaces … source: directory) drifts from the committed truth and produces dual stale installs. Do not use it for testing.
  • Hand-editing .claude/settings.json enabledPlugins desyncs the install records. Use the CLI (claude plugin enable/disable).
  • /reload-plugins is NOT sufficient for newly-added agents/hooks — only a full session restart reliably registers them.

The flow

1. Commit + push + bump version

The installer is version-keyed: install/update no-op if the version already present. So every test cycle bumps the version.

# bump .claude-plugin/plugin.json + .claude-plugin/marketplace.json version (e.g. 4.8.0 -> 4.8.1)
git add -A
git commit -F - <<'MSG'
<type>: <subject>            # clean Conventional Commit — NO leading "@", umlauts correct (ä/ö/ü/ß)

<body>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
MSG
git log -1 --format='[%s]'   # verify subject is clean (no stray @)
git push origin main

2. Register the repo marketplace (once) — point at -internal, mind the name collision

We push to and install from neurawork-git/n8n-autopilot-internal (the dev repo). Its marketplace.json name is n8n-autopilot — the same name as the public release marketplace (neurawork-git/n8n-autopilot). Two marketplaces cannot share a name, so:

# if a `n8n-autopilot` marketplace already points at the PUBLIC repo, remove it first:
claude plugin marketplace list                      # check the Source of `n8n-autopilot`
claude plugin marketplace remove n8n-autopilot      # only if it points at the public repo / a directory

claude plugin marketplace add neurawork-git/n8n-autopilot-internal   # GitHub source = single truth

Never add a directory source for this name (that is the cache-drift trap).

3. Pull the new version + install

claude plugin marketplace update <marketplace-name>     # re-pull the just-pushed commit
claude plugin install n8n-autopilot@<marketplace-name> --scope user
# (or `claude plugin update n8n-autopilot@<marketplace-name>` once installed)

4. Restart the session

Plugins — especially new agents and hooks — register at session start. Restart fully. /reload-plugins does not reliably register newly-added agents.

5. Verify registration (post-restart)

Run the verification probe — a short Workflow that confirms the install is real:

// for each expected agentType, confirm it resolves + that its skills: loaded
phase('Verify')
const AGENTS = ['n8n-researcher','n8n-node-verifier','n8n-comprehender','n8n-author',
                'n8n-validator','n8n-deployer','n8n-tester','n8n-mirror']
const r = await parallel(AGENTS.map(a => () =>
  agent(`Reply with exactly: OK ${a}. Then name one skill you have loaded from your context.`,
        { agentType: `n8n-autopilot:${a}`, label: a }).catch(e => ({ a, error: String(e).slice(0,120) }))))
return r
  • Every agent resolves (no "agent type not found") ⇒ registration clean.
  • Each names a loaded skill ⇒ skills: pass-through works.
  • Also confirm SessionStart hooks fired (look for report-session-env / check-mirror-drift output).

Cleaning up a messy state (one-time)

If dual/stale installs or a directory-pointer marketplace exist:

# uninstall is BLOCKED while the plugin is enabled at PROJECT scope (.claude/settings.json).
# `--scope local` does NOT clear that — use --scope project:
claude plugin disable n8n-autopilot@<mp> --scope project
claude plugin uninstall n8n-autopilot@<mp>                     # removes stale install records (per version)
claude plugin marketplace remove <directory-pointer-name>     # drop any directory pointer / public-repo mp
# then: marketplace add (github -internal) -> install -> restart -> verify

Never resolve a messy state by editing cache files or settings.json by hand — only the CLI (claude plugin disable/enable/uninstall, claude plugin marketplace remove/add).

Definition of done

A plugin change is "tested" only when: pushed to main, installed from the GitHub marketplace at the new version, session restarted, and the verification probe is green (all agents resolve, skills load, hooks fire).

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.