# Validation

> Use this skill whenever MCPMate work needs test, lint, build, Inspector, or release-slice verification. Trigger on implementation, review, bug fixes, PR preparation, regression checks, or any request to validate backend, frontend, desktop, or MCP behavior before reporting completion.

- Skill: `loocor/validation` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add loocor/validation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/loocor/validation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: loocor (https://skillmd.com/u/loocor)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/loocor/validation

---


# Validation

Use this skill to choose the smallest complete validation set for the touched surface, run it, and report the result clearly.

## Goals

- Match validation depth to the actual blast radius.
- Keep backend, frontend, desktop, and MCP verification consistent.
- Distinguish new failures from pre-existing failures instead of skipping checks.

## Rules

1. Do not skip validation because it might fail.
2. If a check fails, classify it as one of:
   - caused by the current change
   - pre-existing
   - unrelated transient/tooling issue
3. Fix current-change failures before reporting done.
4. Call out pre-existing or unrelated failures explicitly in the final report.
5. Use routed HTTP tooling instead of raw `curl` or `wget`.

## When to read more

Read `references/validation-playbook.md` when the task needs:

- the full surface-to-command matrix
- Inspector loop details
- evidence handling and reporting expectations

## Final report shape

Report validation in three parts:

1. What you ran.
2. What passed or failed.
3. What residual risk remains, if any.