Skip to main content
Agent support turns a coding agent into an API-capable collaborator that can scan routes, enrich specs, build flows, and debug failures — all backed by reviewable markdown files in your repo.

Skills

A single markdown playbook that teaches agents the Reqbook spec format, file layout, and how to choose the right approach for each task. All agents get the skill.

Slash commands

Two /rqb commands for Claude Code and Codex CLI: /rqb for creating, enriching, and building flows; /rqb-debug for diagnosing failures.

MCP server

Structured tools for MCP-compatible agents: exec, flow, author, surgical context, and more. Agents get typed results without parsing shell output.

Shared markdown source

Every agent change is a markdown file in api-docs/. Reviewable, diffable, committable. No hidden state.

​
Install in one command

1

Install skill and slash commands

Run the installer from the root of your project. It auto-detects which agents are present and writes the correct files for each.
rqb skills install
The command writes the skill and (for Claude Code / Codex CLI) slash commands to the directories each agent reads.
2

Verify

rqb skills list
Example output:
claude-code detected .claude/skills/rqb/SKILL.md
                      .claude/commands/rqb.md
                      .claude/commands/rqb-debug.md
cursor     detected   .cursor/rules/rqb.mdc
copilot    detected   .github/instructions/rqb.instructions.md
3

Register the MCP server

The MCP server gives compatible agents structured access to spec execution, pipelines, context packs, and authoring — no terminal output parsing.
rqb install mcp
To target one agent:
rqb install mcp --agent=codex-cli
rqb install mcp --agent=cursor
rqb install mcp --agent=copilot
For token-sensitive agent runs, start with rqb_context using mode: "surgical", brief: true, max_fields: 6, and an explicit intent. If an endpoint fails, call rqb_diagnose before reading backend source. Together they return the bounded contract, likely cause, inspect targets, and verify commands before the agent decides how much source inspection is necessary.

​
Per-agent install

rqb skills install --agent=claude-code
# Installs: 1 skill + 2 slash commands
# Locations: .claude/skills/rqb/, .claude/commands/

​
What each agent gets

AgentSkillsSlash commandsMCP
Claude Code1 skill2 commands✓
Codex CLI1 skill2 commands✓
Cursor1 skill✓
GitHub Copilot1 skill✓
Antigravity1 skill✓
OpenCode1 skill✓
Windsurf1 skill✓

​
MCP config locations

AgentCommandConfig
Claude Coderqb install mcp --agent=claude-code.mcp.json
Codex CLI / IDErqb install mcp --agent=codex-cli.codex/config.toml
Cursorrqb install mcp --agent=cursor.cursor/mcp.json
GitHub Copilot in VS Coderqb install mcp --agent=copilot.vscode/mcp.json
OpenCoderqb install mcp --agent=opencodeopencode.json
Antigravityrqb install mcp --agent=antigravity~/.gemini/antigravity/mcp_config.json
Windsurf / Cascaderqb install mcp --agent=windsurf~/.codeium/windsurf/mcp_config.json

​
Keep skills current

After upgrading Reqbook, check whether installed skills match the current binary: 
rqb doctor
To update stale skills automatically: 
rqb doctor --fix
Stale skills are the most common cause of agents missing new commands or writing files in unexpected ways. Run rqb doctor --fix after every Reqbook upgrade.