{
  "markdown": "# AgentKeeper Claude Code Plugin\n\nAgentKeeper provides real-time security scanning, threat detection, and compliance auditing for Claude Code. It works immediately on install — **no account, API key, or configuration required**. Built by [RAD Security](https://rad.security).\n\n## Install\n\n```\n/plugin marketplace add rad-security/claude-code-plugin\n/plugin install agentkeeper\n/reload-plugins\n```\n\nTo update later (third-party marketplaces do not auto-update):\n\n```\n/plugin marketplace update agentkeeper\n```\n\n## What it does\n\n[agentkeeper](./plugins/agentkeeper) hooks into Claude Code at four points — `UserPromptSubmit`, `PreToolUse`, `PostToolUse`, and `SessionStart` — to flag threats before they execute:\n\n- **Real-time threat detection** — credential exfiltration, reverse shells, prompt injection, and 24+ patterns.\n- **`/agentkeeper:audit`** — grades your Claude Code setup for misconfigurations.\n- **`/agentkeeper:inspect`** — audits installed plugins and skills for malicious behavior.\n- **`/agentkeeper:secrets`** — scans your project for exposed keys and credentials.\n- **`/agentkeeper:recap`** / **`/agentkeeper:scan`** — session summary and full host security scan.\n\nDefault mode is **warn** — threats are flagged, not blocked. Switch to blocking with `/agentkeeper:setup`.\n\n## Security & Privacy\n\nThis plugin installs PreToolUse hooks that run local shell scripts. We treat transparency about what they do as a first-class concern:\n\n- **Local-first, fail-open.** In local-only mode (no account connected) the plugin makes **zero network calls and sends zero telemetry** — all detection runs on your machine using a bundled engine. Every hook fails *open*: any error in a hook allows the tool call to proceed, so AgentKeeper can never block your work by failing.\n- **What runs.** Hooks inspect the prompt and tool input (command text, file paths, URLs) in-process to match threat patterns. Session data stays in `~/.agentkeeper-plugin/`.\n- **What is sent, and where — only when connected.** Running `/agentkeeper:connect` links a free account. From then on, hooks call the AgentKeeper API (`agentkeeper.dev`) to use the full pattern engine, org policies, and fleet visibility. Decision metadata for matched tool calls is sent to your dashboard; it is never shared with third parties.\n- **Credential handling.** Connecting provisions a per-device key stored locally under your home directory (`~/.agentkeeper/` / `~/.agentkeeper-plugin/`); it is the only credential the plugin holds and is used solely to authenticate to your own dashboard. Disconnect and remove all hooks/keys any time with `/agentkeeper:disconnect`.\n- **No secret values leave your machine.** The secret scanner reports *locations and types* of exposed credentials — it never transmits or prints the secret values themselves.\n- **License.** MIT — source is fully auditable in this repo.\n\n## Organization Deployment\n\nAdmins can deploy to an entire org via Claude Desktop:\n\n1. Organization settings → Plugins → Connect this GitHub repo\n2. Set AgentKeeper to \"Required\" for automatic deployment\n3. Manage policies centrally at [agentkeeper.dev](https://www.agentkeeper.dev)\n\n## License\n\nMIT — by [RAD Security](https://rad.security)\n",
  "bytes": 3213,
  "sha": "7bf7f8e8363e82649c44c49057c42f2fe2ce8e5c9d0774bafdaf295e6f1d2174",
  "repo_slug": "rad-security/claude-code-plugin",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_rad_security_claude_code_plugin_clawkeep_055f22bb/readme"
}