skill-discovery

Goal

Avoid “missing context” by making skill discovery a deliberate first step.

When to use

Use this whenever the task is:

• Multi-step or multi-file
• Touching an unfamiliar area (API/server, storage, scanner, UI, smoke)
• Debugging CI/smoke failures
• Changing REST contracts

Procedure (fast, repeatable)

1. Identify the domain(s)

• Rust backend: CLI, server, storage, scanner
• Frontend: Vue UI under view/
• Smoke: end-to-end harness under smoke/
• Docs/contracts: docs/openapi.json, docs/help/*

2. Load path-scoped rules

• If you will touch Rust: .github/instructions/backend.instructions.md
• If you will touch UI: .github/instructions/frontend.instructions.md
• If you will touch smoke tests: .github/instructions/smoke.instructions.md

3. Load the 1–3 most relevant skills

Keyword → skill mapping (pick the best match):

• “agent scripts”, “low noise”, “dot reporter”, “no ANSI” → .github/skills/agent-friendly-commands/SKILL.md
• “endpoint”, “REST”, “DTO”, “OpenAPI”, “schema” → .github/skills/api-contract-change-end-to-end/SKILL.md
• “nextest”, “Rust tests”, “integration test failing” → .github/skills/nextest-targeted-testing/SKILL.md
• “Vitest”, “UI tests”, “typecheck”, “frontend test” → .github/skills/vitest-targeted-testing/SKILL.md
• “smoke”, “Playwright”, “E2E”, “serve” → .github/skills/smoke-suite-debugging/SKILL.md
• “local dev”, “Vite”, “lotar serve”, “ports”, “SSE” → .github/skills/local-dev-serve-troubleshooting/SKILL.md
• “task tracking”, “.tasks”, “LoTaR tasks” → .github/skills/lotar-task-tracking/SKILL.md

4. Verify before you assume

• Confirm current behavior (tests/repro) before changing semantics.
• Search for existing helpers before introducing new ones.

Safety note:

• AGENTS.md is the canonical policy (tests required, no secrets/PII, no git operations for recovery). Skills only add extra context.

Continuous improvement

If you find yourself looking up the same thing twice (commands, env vars, gotchas), add it to the most relevant skill or create a new narrowly-scoped one. Keep changes minimal and link out to docs instead of duplicating them.