agent-skills: Instructions file for Codex

Contributor Guide

This repository contains a collection of agent skills. Follow these guidelines when contributing.

Repository Structure

skills/
└── <skill-name>/
    ├── SKILL.md           # Required: skill definition with frontmatter
    ├── api-reference/     # Optional: API documentation files
    ├── guides/            # Optional: usage guides
    ├── rules/             # Optional: rule files for compiled skills
    └── metadata.json      # Optional: version and metadata

Creating a New Skill

1. Create a new directory under skills/ with a descriptive kebab-case name
2. Add a SKILL.md file with the required frontmatter:

---
name: my-skill-name
description: >-
  A concise description of what the skill does and when to use it.
---

1. Add supporting files as needed (API references, guides, rules)

SKILL.md Requirements

• Must include name and description in YAML frontmatter
• Description should clearly state when the skill should be activated
• Instructions should be concise and action-oriented
• Reference supporting files using relative paths

Writing Style for Skill Content

These files are agent skills, not CLI manuals. The reader is an AI agent that can always run teable <cmd> --help at runtime, so:

• Never restate what --help already says — no flag inventories, no per-command syntax walkthroughs, no one-section-per-command structure.
• Write the operational knowledge help cannot provide: when to use which command, decision rules ("if you need X, use A; otherwise B"), cross-command workflows, prerequisites, pitfalls, value formats, safe defaults, and warnings for destructive operations.
• Prefer pitfall tables and decision rules over command catalogs (see the Common Pitfalls table in cli-reference.md for the house style).
• Examples only where the invocation shape is non-obvious — complex JSON payloads, quoting, multi-step sequences. Trivial invocations need none.
• Brevity is a feature: skill files are loaded into an agent's context window; every line costs tokens on every task that uses the skill.
• The reader runs outside Teable-managed sandboxes: this skill is consumed by agents operating @teable/cli in their own environment, never inside a Teable-managed agent sandbox. The manifest's availability markers (managed-sandbox-disabled, sandbox-only) describe runtime gating those readers never encounter — do not thread sandbox-conditional wording through the docs, and do not document sandbox-only command groups (e.g. sandbox tool).