skill-discovery

localtaskrepo's avatarfrom localtaskrepo

Use this at the start of non-trivial work to quickly load the right runbooks (skills + path-scoped instructions) before making changes.

⭐1stars🔀0forks📁View on GitHub🕐Updated Jan 7, 2026
[Memory and Context Management]

When & Why to Use This Skill

The Skill Discovery tool is a specialized Claude skill designed to optimize agentic workflows by proactively loading relevant runbooks, path-scoped instructions, and domain-specific skills. By making context retrieval a deliberate first step, it eliminates the 'missing context' problem, ensuring the AI agent operates with full awareness of project-specific rules, architectural patterns, and testing protocols before executing non-trivial changes.

Use Cases

  • Onboarding to Unfamiliar Codebases: Automatically identifying and loading backend (Rust), frontend (Vue), or smoke test instructions when navigating new parts of a repository.
  • Complex Debugging Sessions: Streamlining the resolution of CI/smoke failures by pulling in targeted testing skills like Vitest, Nextest, or Playwright configurations.
  • API and Schema Updates: Ensuring REST contract guidelines and OpenAPI documentation are prioritized when modifying service endpoints or DTOs.
  • Standardizing Development Workflows: Using keyword-to-skill mapping to quickly activate task tracking, local dev troubleshooting, or agent-friendly command sets.
  • Multi-step Engineering Tasks: Preventing errors in multi-file refactoring by loading the correct architectural 'runbooks' before making any semantic changes.
nameskill-discovery
descriptionUse this at the start of non-trivial work to quickly load the right runbooks (skills + path-scoped instructions) before making changes.

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/*
  1. 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
  1. 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
  1. 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.