AGENT SKILL

using-agent-skills

Selects and applies the smallest relevant set of skills for a task, preventing generic prompting and unused specialist guidance. Use at the start of every command, task, review, or repair cycle.

License 许可
Apache-2.0
最近更新
2026-07-13
Stars
★ 0
Forks
0

01 前言 Front Matter

name
using-agent-skills
description
Selects and applies the smallest relevant set of skills for a task, preventing generic prompting and unused specialist guidance. Use at the start of every command, task, review, or repair cycle.

03 SKILL.md 正文

Using Agent Skills

Purpose

Ensure the agent deliberately activates the right skills instead of relying on a single broad instruction. The harness should be skill-routed, not vibes-routed.

Required workflow

  1. Identify the lifecycle stage: intent, context, expectations, task derivation, implementation, review, validation, commit, reporting, delivery, teaching.
  2. Select only the skills that materially affect the next action.
  3. Read each selected skill's trigger, constraints, red flags, and verification checklist.
  4. State the selected skills in the action record or result artifact.
  5. Do not invoke project-pack skills unless the configured project pack is active.
  6. Do not use a domain skill as a substitute for generic test, engineering, or validator review.

Skill routing baseline

Stage Required skills
Intent intent-capture
Context context-assembly, environment-bootstrap
Expectations expectation-derivation, expectation-validation
Task derivation epic-task-decomposition
Build red-green-vertical-slice, simple-python-implementation, testing-patterns
Review code-review, gilfoyle-code-review when high scepticism is needed
Python design simple-python-implementation, oop-design-patterns only when OO is justified
Comments/docs self-explanatory-code-comments
Containers docker-containerization
CI/CD github-actions-ci-cd
Azure IaC azure-naming, bicep-best-practices
Commit verified-commit
PR pull-request-delivery
Session report session-reporting
Teaching teach-me

Red flags

  • A command proceeds without naming any active skill.
  • The agent uses every skill, creating context noise.
  • A specialised skill is used outside its domain.
  • A reviewer uses implementation skills to modify code.
  • The builder sees validator-only hidden checks.

Verification

  • [ ] Skill selection is recorded.
  • [ ] Selected skills match the lifecycle stage.
  • [ ] Domain skills match the active project pack.
  • [ ] No irrelevant high-token skill was loaded.

04 同仓库更多 Skills