<aside> 🚧

Under Construction. This page is currently incomplete and missing significant content across multiple sections.

</aside>

Agents

Agents are system prompts + agent settings in the provider (name, model, when it is activated, allowed tools, skills, among others)

Comparison table between providers

GitHub Copilot Claude Code Codex CLI OpenClaw
Directory .github/agents .claude/agents n/a n/a (JSON config)
Where they are located Files *.agent.md frontmatter *.md frontmatter files In [agents] section in .codex/config.toml agents.list[] in openclaw.json
name / id name (optional, defaults to filename) name (required) [agents.<n>] (TOML section name) id (required, unique)
description description (required, 1-500 chars) description (required) description (optional) name (display name, optional)
model model (optional) model (sonnet/opus/haiku/inherit) model in config_file model.primary • model.fallbacks
tools allowlist tools (list or ["*"]) tools (comma-separated, inherits if omitted) ❌ (inherited from global config) tools.allow (list)
tools denylist ❌ (only allowlist) disallowedTools ❌ tools.deny (list)
tools by provider ❌ ❌ ❌ tools.byProvider (per-model restrictions)
MCP servers mcp-servers (org/enterprise only) mcpServers ❌ (global MCP config only) ❌ (global MCP config only)
skills ❌ skills (comma-separated) ❌ Via workspace skills/ folder
memory ❌ memory (user/project) ❌ Per-agent workspace isolation
permission mode ❌ permissionMode (default/acceptEdits/bypassPermissions/plan) approval_policy in config_file ❌ (global only)
sandbox ❌ ❌ sandbox_mode in config_file sandbox.mode (off/non-main/all), sandbox.scope (session/agent/shared)
workspace ❌ (implicit) ❌ (implicit) ❌ (global) workspace (path, required)
routing / bindings ❌ ❌ ❌ bindings[] (match by channel/peer/group)
default agent ❌ ❌ ❌ default: true
hooks ❌ hooks ❌ Via workspace hooks/ folder
max turns ❌ maxTurns ❌ ❌
target environment target (vscode/github-copilot) ❌ ❌ ❌
handoffs handoffs (VS Code only) ❌ ❌ ❌ (via sessions_spawn tool)
auto-infer infer (boolean, defaults to true) ❌ (always by description) ❌ ❌ (routing via bindings)
config file reference ❌ ❌ config_file (path to TOML) ❌ (inline config)
model reasoning effort ❌ ❌ model_reasoning_effort in config_file ❌
system prompt Markdown body Markdown body developer_instructions in config_file Via workspace SOUL.md • AGENTS.md
Ref (link) GitHub Docs Claude Code Docs Multi-agents Multi-Agent Sandbox & Tools

Instructions

*Instructions are system prompts that can be injected into the system prompt of a session. There are two types of Instructions:

Comparison table

GitHub Copilot Claude Code Codex CLI
Always-on file .github/copilot-instructions.md
or AGENTS.md
or CLAUDE.md CLAUDE.md AGENTS.md
Directory .github/instructions .claude/rules/ ❌
Files *.instructions.md frontmatter *.md frontmatter Configurable at the CLI level in project_doc_fallback_filenames
GLOB filter applyTo paths ❌
Ref (link) https://code.visualstudio.com/docs/copilot/customization/custom-instructions https://code.claude.com/docs/en/memory https://developers.openai.com/codex/guides/agents-md#customize-fallback-filenames

Prompts

Prompts are user prompt templates that can be invoked/applied in a session using slash commands and can include variables, including responses from the user themselves.

Comparison table

GitHub Copilot Claude Code Codex CLI
Feature available? ✅ ❌ Only using user-invocable Skills ❌
Directory .github/prompts
Files *.prompts.md frontmatter
Ref (link) https://code.visualstudio.com/docs/copilot/customization/prompt-files https://code.claude.com/docs/en/skills#control-who-invokes-a-skill

Skills

Skills are complete sets of Agents and/or Tools and/or Instructions and/or Scripts, usually in the openskills.io format.