.github/skills/agent-file-specs/SKILL.md
Contains the complete specifications for AI coding assistant customization files including agents, skills, prompts, and instructions. Works with GitHub Copilot, Claude Code, Codex, OpenCode, and other providers. Use this skill when you need to reference the correct file format, required fields, supported attributes, file locations, or VS Code settings for any customization file. Follows the Agent Skills open standard (agentskills.io).
npx skillsauth add jburlison/metaprompts agent-file-specsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
This skill provides the authoritative, up-to-date specifications for all AI coding assistant customization file types. Each file type has its own dedicated reference document with complete attribute tables, validation rules, examples, and best practices.
Standard: These specs align with the Agent Skills open standard and the VS Code Copilot customization documentation.
Important: Detailed specifications live in the
references/folder — not in this file. This keeps the main skill file lightweight for efficient context loading per the progressive disclosure model.
| Type | Extension | Purpose | Reference |
|------|-----------|---------|-----------|
| Agent | .agent.md / .md | Custom AI personas with specialized behaviors, tool restrictions, and handoff workflows | AGENTS.md |
| Skill | SKILL.md (directory-based) | Reusable, portable capabilities with scripts and resources — follows the Agent Skills open standard | SKILLS.md |
| Prompt | .prompt.md | Reusable prompt templates invoked as slash commands | PROMPTS.md |
| Instruction | .instructions.md | Contextual guidance applied automatically by file type or description | INSTRUCTIONS.md |
| File Locations | — | Provider folder mapping, directory structure, naming conventions | FILE-LOCATIONS.md |
| VS Code Settings | — | Settings that control customization file discovery and behavior | SETTINGS.md |
These specifications work across multiple AI coding assistant providers. Each provider uses its own base folder but the file formats are consistent:
| Provider | Base Folder |
|----------|-------------|
| GitHub Copilot | .github/ |
| Claude Code | .claude/ |
| Codex | .codex/ |
| OpenCode | .config/opencode/ |
See FILE-LOCATIONS.md for the complete directory structure and provider-specific details.
Use this decision tree to choose the right file type:
"I want coding standards applied everywhere" → Use always-on instructions
(.github/copilot-instructions.md or AGENTS.md).
See INSTRUCTIONS.md.
"I want different rules for different file types" → Use file-based
instructions (.instructions.md with applyTo patterns).
See INSTRUCTIONS.md.
"I have a reusable task I run repeatedly" → Use a prompt file
(.prompt.md) invoked as a slash command.
See PROMPTS.md.
"I need a multi-step workflow with scripts and resources" → Use an Agent
Skill (directory with SKILL.md). Portable across tools.
See SKILLS.md.
"I need a specialized AI persona with tool restrictions" → Use a custom
agent (.agent.md). Supports handoffs for multi-step workflows.
See AGENTS.md.
development
Validates AI coding assistant customization files (agents, skills, prompts, instructions) for correct format and structure. Works with GitHub Copilot, Claude Code, Codex, OpenCode, and other providers. Use when checking if agent files are properly configured, troubleshooting agent issues, or before committing new customization files.
development
Generates documentation and usage guides for agents, skills, prompts, and instructions. Works with GitHub Copilot, Claude Code, Codex, OpenCode, and other providers. Use when onboarding team members, creating README files for your customizations, or generating usage examples for existing agents.
development
Analyzes existing agents, skills, prompts, and instructions to identify overlaps, redundancies, and conflicts. Works with GitHub Copilot, Claude Code, Codex, OpenCode, and other providers. Use before creating new customization files to avoid duplication, when consolidating agents, or when troubleshooting conflicting behaviors.
development
Maintainer-only workflow for handling GitHub Secret Scanning alerts on OpenClaw. Use when Codex needs to triage, redact, clean up, and resolve secret leakage found in issue comments, issue bodies, PR comments, or other GitHub content.