Mode d’emploi
- Copiez la commande et examinez le paquet avant de l’installer dans votre environnement agent.
- Consultez le document original pour les outils, identifiants et environnements requis.
- Décrivez les entrées et le livrable souhaité ; comparez la sortie aux sources.
npx skills add https://github.com/legout/skills --skill effective-agent-skillsAvant de commencer
- L’installation n’inclut ni abonnement au modèle, ni quota API, ni accès aux fichiers.
- Étoiles et installations reflètent l’activité, pas un test indépendant de qualité ou de sécurité.
Document Skill original
Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.
Adapted from David Ondrej's `effective-agent-skills` at commit41cb3f03deb34567cd0d3d5eae016e6408cce1fa(MIT). Recurring-correction enforcement is adapted from Cursor's `poteto-mode` at commit93b00b89ef425a9c1bac0d0b317dfc49c930ac99(MIT, Copyright (c) 2026 Cursor).
Effective Agent Skills
Workflow
- Define one coherent capability or discipline.
- Write the description first: capability and concrete triggers in one or two short sentences, important terms first. Add a boundary only to distinguish nearby skills; keep workflow steps in the body.
- Keep the ordinary execution path, safety gates, output contract, and verification loop in
SKILL.md. - Move uncommon detail into directly linked
references/; put fragile or repetitive logic in testedscripts/. - Test positive triggers, near-miss negatives, explicit invocation, execution, failure handling, and links.
Read the authoring guide for anatomy, design patterns, composition, security, and detailed examples.
Rules
- Use valid YAML frontmatter with
nameand one concisedescriptionline. - Match the skill name to its leaf directory and keep names stable.
- Use relative paths and one reference level; tell the agent exactly when to open each file.
- Do not reteach general knowledge, bundle unrelated workflows, or include human-facing README files.
- Preserve approval, publication, credential, destructive-action, and evidence-before-claim gates.
- Prefer installed tools and
--helpover copied command catalogs. - When the same correction recurs, decide whether it belongs in concise prose, routing metadata, an evaluation, or executable enforcement. Prefer a lint rule, schema, metadata constraint, test, or script when prose repeatedly fails; do not automate subjective preferences that still require judgment.
- Treat skills as code: validate, test, review, and version them.
Verification
Confirm frontmatter parses strictly, every local link resolves, scripts run, the skill triggers on intended prompts, nearby skills win their own prompts, and the shortest supported workflow still completes. Test description routing separately from workflow execution. Inspect actual SKILL.md loads, not the agent's self-report; check discovery and invocation settings before blaming the description.
