skills-ttt
par chengxindl
Coding Agents & IDEsRésumé
Guide for creating effective skills that extend agent capabilities with specialized knowledge, workflows, or tool.
Description
Ce que fait ce skill
A guide for creating modular skill packages that extend agent capabilities with specialized knowledge, workflows, and tool integrations. Covers skill structure (SKILL.md frontmatter, scripts, references, assets), design principles like progressive disclosure, and a six-step creation process from gathering requirements to iteration.
The three-level loading system (metadata, then SKILL.md body, then bundled resources) keeps context window usage proportional to task complexity rather than loading everything upfront.
Quand l'utiliser
- Scaffolding a new skill directory for a domain-specific workflow
- Adding reusable scripts to avoid rewriting PDF or image processing code each session
- Reorganizing an overgrown skill into separate reference files by domain or variant
- Writing trigger descriptions so the right skill activates on the right user request
- Validating skill naming conventions and frontmatter before publishing
Exemple de workflow
User asks: Help me create a skill for querying our BigQuery tables
- 1
Asks clarifying questions to establish concrete usage examples and which table domains matter
- 2
Identifies references/schema.md as a reusable resource to document table schemas and relationships
- 3
Runs init_skill.py to scaffold the skill directory at ~/.deepagents/agent/skills/bigquery
- 4
Writes SKILL.md frontmatter with comprehensive trigger descriptions and body instructions referencing schema.md
- 5
Runs quick_validate.py to confirm required fields, naming conventions, and description length
A bigquery skill directory with SKILL.md and references/schema.md ready for agent use
Installation
clawhub install chengxindl/skills-tttExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.