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
- dzianisv/opencode-pluginstheir own version, first seen Nov 1, 2026
- glittercowboy/plugin-freedom-systemtheir own version, first seen Oct 8, 2026
- polyipseity/obsidian-terminaltheir own version, first seen Oct 8, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- polyipseity/obsidian-terminal★ 960same name
- glittercowboy/plugin-freedom-system★ 217same name
- dzianisv/opencode-plugins★ 11same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository25 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- build-stack-v2Skill
- build-workflowSkill
- build-workflow-v2Skill
- check-mcpsSkill
- data-tablesSkill
- deploySkill
- feedbackSkill
- find-credentialSkill
- find-projectSkill
- init-repoSkill
- inventorySkill
- mirror-syncSkill
- n8nac-cheatsheetSkill
- n8nac-referenceSkill
- n8n-code-javascriptSkill
- n8n-code-pythonSkill
- n8n-modeSkill
- n8n-orchestration-patternsSkill
- n8n-structured-extractionSkill
- pull-schemasSkill
- session-envSkill
- stack-intakeSkill
- sync-credentialsSkill
- test-manualSkill
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
- Very specific: line 58
Never add a directory source for this name (that is the cache-drift trap).
- Distinctive: line 72
Run the verification probe — a short Workflow that confirms the install is real:
- Distinctive: line 83
Every agent resolves (no "agent type not found") ⇒ registration clean.
- Distinctive: line 88
If dual/stale installs or a directory-pointer marketplace exist:
- Very specific: line 97
Never resolve a messy state by editing cache files or settings.json by hand — only the CLI (claude plugin disable/enable/uninstall, claude …
“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.
Files
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-pluginsdoes 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.jsonenabledPluginsdesyncs the install records. Use the CLI (claude plugin enable/disable). /reload-pluginsis 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 main2. 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 truthNever 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-driftoutput).
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 -> verifyNever 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).
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
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.

