agent-skills: Instructions file for Codex

AGENTS.md

Purpose

Agent Skills repository in Agent Skills format.

How to use this repository

• Skills are located at skills/

• Agent-visible paths:

  • .agents/skills/
  • .claude/skills/

• Setup virtual env via uv

• Use uv run agentskills-ci score --min-score 80 ./skills to validate skills

• Use uv run mdlint check skills/**/*.md to validate skill's markdown files

• Use ruff, zuban, pytest to lint/test skill's scripts

• Skills are reusable and independent

When solving tasks:

1. Identify if a relevant skill exists
2. Prefer using a skill over improvising
3. Follow the skill strictly once selected

Output quality

Always:

• prioritize correctness over verbosity
• surface risks and uncertainties explicitly
• structure outputs (sections, bullet points)
• avoid vague statements

Safety and scope

• Do not assume access to external systems unless explicitly provided
• Do not fabricate data (PRs, logs, metrics, etc.)
• Clearly state assumptions

Modification rules

When editing this repository:

• keep skills portable (vendor-neutral)
• do not introduce agent-specific logic into SKILL.md
• place agent-specific config under skills/<name>/agents/
• keep .agents/skills/ and .claude/skills/ symlinks in sync with skills/
• prefer repo-provided tooling from the uv run

Adding new skills

When creating a skill:

• ensure description is an activation rule
• include:
  • "When to use"
  • "Steps"
  • "Output"
• avoid duplication with existing skills
• consult the Agent Skills references instead of duplicating their guidance here:
  • Specification: https://agentskills.io/specification
  • Optimizing descriptions: https://agentskills.io/skill-creation/optimizing-descriptions
  • Best practices: https://agentskills.io/skill-creation/best-practices
• use repository tooling where applicable:
  • git hooks for baseline checks
  • agentskills CLI from the dev environment