Current DashClaw integration artifacts in one place. Skills ship as zips you can drop into ~/.claude/skills/; plugins, hooks, and the MCP server install from a single command against the source repo.
Skill bundles
Self-contained skill directories Claude Code, Claude Desktop, Codex, and Hermes Agent can load directly. Each zip contains a SKILL.md plus any references and scripts the skill needs.
SKILL.md + references
Teaches an agent how to use DashClaw correctly: risk thresholds, decision handling (allow / warn / block / require_approval), action recording, session lifecycle, plus six new sections for handoffs, secret hygiene, skill safety, open loops, learning, and in-session retrospection.
unzip dashclaw-governance.zip -d ~/.claude/skills/
Auto-installed when you install any of the agent plugins below. Source mirrors live in public/downloads/ and plugins/dashclaw/skills/ in the repo.
Agent plugins
One DashClaw plugin source, three ecosystems. Each plugin ships the MCP server config, both skills above, and an agent identity used to separate sessions by host (claude-code, codex, hermes).
manifests + skills + MCP configs
Full plugin bundle in one zip: the three plugin manifests (Claude Code / Codex / Hermes), MCP configs, both skills, assets, and PLUGIN_PARITY.md. Drop into your agent's plugin directory or extract for inspection. Manifest version v3.3.0.
unzip dashclaw-governance-plugin.zip # produces a dashclaw/ tree you can drop into ~/.claude/plugins/, ~/.codex/plugins/, etc.
plugins/dashclaw/.claude-plugin/plugin.jsonMCP server + governance skill, plus a hooks installer for PreToolUse / PostToolUse / Stop guards over Bash, Edit, Write, and MultiEdit.
# Native plugin marketplace (inside Claude Code): /plugin marketplace add ucsandman/DashClaw /plugin install dashclaw@dashclaw # Or via the CLI, no clone required: npm i -g @dashclaw/cli dashclaw install claude # prompts for endpoint + API key dashclaw install claude --trial # hosted signup, paste the key # From a repo checkout: npm run hooks:install
plugins/dashclaw/.codex-plugin/plugin.jsonInstalls the supported DashClaw controls for Codex in ~/.codex/config.toml: MCP server config, PreToolUse / PostToolUse / Stop hooks, and the governance protocol in AGENTS.md. Idempotent; re-run after upgrades.
node cli/bin/dashclaw.js install codex \ --project /path/to/your/project # Optional: opt in to legacy notify config node cli/bin/dashclaw.js install codex \ --project /path/to/your/project --include-notify
plugins/dashclaw/.hermes-plugin/plugin.yamlEight lifecycle hooks: pre/post tool, pre/post LLM call with per-turn governance context injection, on-session start/end with live ingest finalize, secret redaction in tool output, and subagent_stop ROI tracking. Wires handoffs end-to-end across sessions.
# macOS / Linux bash scripts/install-hermes-plugin.sh # Windows powershell -File scripts/install-hermes-plugin.ps1 # Sanity check hermes dashclaw doctor
Model Context Protocol
17 governance tools plus 3 read-only resources over MCP. The on-disk server ships with the plugins above. Deployments can also expose Streamable HTTP at /api/mcp when the route is enabled and reachable; authentication is required. MCP calls remain cooperative unless an installed hook or supported gateway mechanically enforces them.
{
"mcpServers": {
"dashclaw": {
"command": "node",
"args": [
"/path/to/DashClaw/mcp-server/bin/dashclaw-mcp.js",
"--agent-id", "claude-code"
],
"env": {
"DASHCLAW_URL": "https://your-instance.vercel.app",
"DASHCLAW_API_KEY": "oc_live_..."
}
}
}
}mcp_servers=[{
"type": "url",
"url": "https://your-instance.vercel.app/api/mcp",
"headers": {"x-api-key": "oc_live_..."},
"name": "dashclaw"
}]Full tool catalogue and resource list: /docs#mcp-server.
Claude Code hooks
Govern configured Claude Code tool categories without per-call SDK code. Installs PreToolUse, PostToolUse, Stop, and SessionStart hooks plus the tool-classification module into .claude/hooks/, then merges the relevant blocks into .claude/settings.json. Idempotent; re-run after upgrades.
hooks + agent_intel + tests
The hook scripts (pretool, posttool, stop, and the enforcement-liveness probe), the dashclaw_agent_intel/ tool-classification module, default settings.json, and the test suite. Drop the unzipped hooks/ directory into your project's .claude/hooks/.
unzip dashclaw-claude-code-hooks.zip -d <your-project>/.claude/ # Then merge .claude/hooks/settings.json snippets into your .claude/settings.json
npm run hooks:install
node /path/to/DashClaw/scripts/install-hooks.mjs --target=.
The Stop hook reports token usage observed in the session transcript and attaches it to action records opened by the pretool during that turn. The bundled liveness probe is also a point-in-time client report, not continuous attestation. Required env: DASHCLAW_BASE_URL, DASHCLAW_API_KEY, optional DASHCLAW_HOOK_MODE=enforce.
Programmatic surface
For custom agents and frameworks. Bare SDK calls are cooperative. The runGoverned and run_governed helpers require a protocol-1-compatible server before they invoke a callback, and uncertain claims or outcomes must be reconciled before retry. Versions current as of this build.
Canonical 41-method surface for core governance, protocol-1 bounded invocation, execution outcomes, security scanning, sessions and action graphs, agent identity, risk signals, policy simulation, plan authorization, containment, and team tasks.
npm install dashclaw
Broader Python surface (61 methods) with protocol-1 bounded invocation and framework integrations for CrewAI task instrumentation and AutoGen conversation monitoring.
pip install dashclaw
Full integration reference, API method tables, OpenAPI spec, and self-host runbook live in /docs. Source code on GitHub.