agentleFS
Sign inStart free
fruggrfruggr/zendesk-mcp-server/.claude/skills/functional-testing/SKILL.md

functional-testing

Skill15 starsChanged 20 days agoMITView on GitHub

Drive the inter-LLM functional test harness for this MCP server. Use when the user invokes /functional-testing — either with a scenario slug (e.g. `01-namespace-readonly`) or `all`. Reads STATE.md, produces a bridge prompt for the local executor, then ingests reports and writes verdicts. Never reads `expected.md` until the executor has already pushed its report.

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. mejbaurbahar/fagun★ 1same name
    2. mejbaurbahar/software-tester-skills★ 1same 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 repository5 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

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

    Refuse if not your turn: If holder in the frontmatter is not leading, the executor still owes a push.

    line 37
  2. Distinctive:

    Wait for the user to confirm the push: No polling. The user signals you in chat.

    line 61
  3. Very specific:

    Never read expected.md before the executor has pushed their report.

    line 109
  4. Very specific:

    Never edit spec.md or expected.md mid-run.

    line 110
  5. Distinctive:

    Never push a code fix and the verdict in the same commit.

    line 111
  6. Very specific:

    If you can't reach a verdict (raw JSON corrupted, missing artifact), say so in verdict.md and ask the user — don't fabricate.

    line 112

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Functional testing — leading LLM protocol

You are the leading LLM in a two-LLM functional test loop. The other LLM runs locally on the user's machine (Claude Code CLI), executes scenarios against the real MCP server build, and pushes its artifacts. You drive the loop without ever running the server yourself.

Invocation forms

  • /functional-testing <NN-slug> — drive a single scenario.
  • /functional-testing all — drive every scenario in STATE.md whose status is not yet final (i.e. not OK and not FAIL), in order.

Scenario status lifecycle

STATE.md rows go through three states:

statuswritten bymeaning
pendinginitialscenario declared, executor has not yet run it
doneexecutorreport pushed, awaiting leading-LLM verdict
OK/FAILleading LLMverdict written; row is now final

/functional-testing all picks the first row whose status is pending or done (everything that hasn't reached a final verdict).

Loop per scenario

  1. Read state. Open tests/functional/STATE.md. Confirm the scenario exists and its status. If the user passed all, pick the first row whose status is not yet final (still pending or done).
  2. Refuse if not your turn. If holder in the frontmatter is not leading, the executor still owes a push. Tell the user, stop.
  3. Read the spec, NOT the expected. Open tests/functional/scenarios/<slug>/spec.md. Do NOT open expected.md yet — opening it before the executor reports would let bias leak into the bridge prompt.

  4. Produce a bridge prompt for the user to paste into their local Claude Code session. Template:

    Tu es dans le repo zendesk-mcp-server, branche locale doit être à jour
avec origin/<branch-from-STATE>.
Lis tests/functional/scenarios/<slug>/spec.md et applique strictement
ses instructions.
Ne lis PAS expected.md — c'est le critère de verdict, le voir biaiserait
ton rapport.
Écris tests/functional/reports/<slug>.raw.json + tests/functional/reports/<slug>.report.md.
Quand fini, mets à jour tests/functional/STATE.md (status=done,
holder=leading), commit et push.
Préviens-moi.

    Adapt the language to the user's (AGENTS.md says GitHub = English, chat = user's language).

  5. Wait for the user to confirm the push. No polling. The user signals you in chat.
  6. Pull and read. git pull origin <branch>. Open the executor's artifacts: tests/functional/reports/<slug>.raw.json (canonical, ground truth) and tests/functional/reports/<slug>.report.md (executor's narrative + assertion fence).
  7. Now read expected.md. Open tests/functional/scenarios/<slug>/expected.md. Compare each assertion ID between the report fence and the expected values. Re-verify each against raw.json — don't trust the executor's actual blindly; they may have miscopied.

  8. Write the verdict at tests/functional/reports/<slug>.verdict.md:

    # Verdict — <slug>

| ID  | Status | Notes                                                  |
| --- | ------ | ------------------------------------------------------ |
| A1  | OK     | -                                                      |
| A2  | FAIL   | Expected X, raw shows Y. See `raw.json` `.tools[1]`.   |

## Summary

<green / list of failing IDs>

## Proposed actions

- <if any FAIL: pointer to the code fix, file:line>
  9. Update STATE.md. Set the scenario status to OK or FAIL — this is the final state, overwriting the done that the executor wrote. Bump holder back to executor only if you need them to re-run (e.g. report was incomplete); in that case revert status to pending. Otherwise leave holder: leading and consider the scenario closed.
  10. Commit and push the verdict and STATE update.
  11. If FAIL with a clear root cause: propose the code fix on the branch in a separate commit (don't conflate harness output with code changes). Push, mention to the user.

When all

Loop steps 1-11 for each pending scenario, in order. Stop on the first FAIL and surface the diagnosis to the user before continuing — the executor may need to re-run after a code fix.

Hard rules

  • Never read expected.md before the executor has pushed their report.
  • Never edit spec.md or expected.md mid-run.
  • Never push a code fix and the verdict in the same commit.
  • If you can't reach a verdict (raw JSON corrupted, missing artifact), say so in verdict.md and ask the user — don't fabricate.

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.

Check your setup, free