agentleFS
CristhianzlCristhianzl/pocket-claude/.claude/skills/validating-in-reality/SKILL.md

validating-in-reality

Skill2 starsChanged 3 months agoMITView on GitHub

Validate a bugfix or feature against the REAL running system — execute the user's cURL, assert actual responses, check persisted database state, and drive the UI end-to-end — instead of assuming the code change works. Use whenever the user provides a cURL command, a localhost/endpoint URL, or asks to "validate", "test it for real", "check the database", "no assumptions", or when finishing a bugfix/feature in a project with a running server. The user's cURL IS the acceptance test. Not a substitute for writing-tests (automated coverage) or /verify (static gates) — this is live-system proof on top of them.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep validation read-only beyond the tested operation: don't mutate unrelated data.

    line 34
  2. Very specific:

    If you discover how this project runs/seeds/authenticates for validation (base URL, test users, DB access command), append a …

    line 39

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Validating in reality

A change isn't done because the code looks right — it's done when the running system proves it. This skill turns a user-provided cURL (or endpoint, or repro) into the acceptance test, and replaces "should work" with pasted evidence.

Read first (always)

List learnings/ and read anything relevant — how this project starts its server, seeds data, connects to the DB, and authenticates test requests lives there and overrides the defaults here.

The principle: evidence over inference

Reading the diff tells you what the code should do. Only the running system tells you what it does. When the user supplies a cURL, an endpoint, or a repro, that artifact is the definition of done — the task is complete when it behaves correctly against the live system, and you have the output to show it.

Workflow

  1. Confirm the system is running. Find the server (the project's run command, docker compose ps, the port in the cURL). If nothing is running and you can't start it with the project's documented command, say so explicitly and stop — never simulate a validation you didn't run.
  2. Run the acceptance test BEFORE the change (bug: prove it reproduces — the live RED, mirroring fixing-bugs; feature: prove the behavior is missing). Paste the actual response.
  3. Implement (per fixing-bugs / developing-features — tests included as usual).
  4. Run the acceptance test AFTER. Paste the actual response and diff it against the BEFORE. Status code, body, and headers must show the expected change — actually compare them, don't eyeball.
  5. Check persisted state, not just the response. Query the database (the project's client/ORM shell) and confirm the row/document actually changed — created, updated, deleted — as claimed. An API can return 200 and persist nothing.
  6. Exercise the negative cases: wrong/missing auth, invalid payload, nonexistent ID. The fix must not have opened a hole the happy-path cURL can't see.
  7. If the flow has a UI, drive it end-to-end with skills/playwright-cli: perform the user action in the browser and verify the visible result — the backend being right doesn't prove the frontend wired it.
  8. Report with evidence. For each step: the command run and the real output (trimmed to the relevant part). Redact tokens/secrets/PII from anything you paste.

Rules

  • Never claim "validated" for anything you didn't execute. Partial validation is fine if labeled ("request + DB checked; UI not exercised — no frontend in this flow").
  • The user's cURL is run as given (plus auth fixes they approve) — don't silently substitute a different endpoint that happens to pass.
  • Keep validation read-only beyond the tested operation: don't mutate unrelated data; use test/seed data where the project provides it.
  • These checks complement — never replace — automated tests (writing-tests) and the static gates (/verify).

Capture a learning

If you discover how this project runs/seeds/authenticates for validation (base URL, test users, DB access command), append a learnings/YYYY-MM-DD-slug.md (or use /learn) so the next validation is instant.

See also

  • skills/fixing-bugs — the test-level RED→GREEN this skill mirrors at runtime level.
  • skills/playwright-cli — driving the UI end-to-end.
  • /verify command — static definition-of-done gates (build, types, lint, tests, secrets, diff).
  • hooks/check-real-validation.py — auto-injects this requirement when your request contains a cURL or local endpoint.

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.