acp-agent-skill

An Agent Skill for the Agentic Control Plane (ACP) — see, price, and control every tool call your AI agents make.

Install this skill into any harness that supports the SKILL.md format and your agent knows how to: check whether ACP is present, set it up (one-command local install, no-signup sandbox provisioning, or MCP connection), check sensitive actions before running them, mint scoped child keys for subagents, and propose policy rules that a human confirms in the console.

Install

Claude Code (personal, all projects):

git clone https://github.com/agentic-control-plane/acp-agent-skill ~/.claude/skills/agentic-control-plane

Claude Code (one project): clone into .claude/skills/agentic-control-plane inside the project.

Other harnesses: point your skill loader at SKILL.md; the references/ files are progressive-disclosure detail.

Contents

• SKILL.md — the skill: presence check, setup by situation, acp_check before blast-radius actions, child keys for subagents, human-confirmed policy proposals
• references/setup.md — the four setup paths in detail, credential storage, verification, the nine MCP tools
• references/child-keys.md — POST /api/v1/keys/child request/response shape, error codes, delegation rules

Links

• Site: https://agenticcontrolplane.com · agent-facing surface: /AGENTS.md
• Console: https://cloud.agenticcontrolplane.com · API: https://api.agenticcontrolplane.com
• Open core (MIT): GatewayStack

License

MIT