Get Started with Critiqor: Monitor OpenClaw in Four Steps

Monitor your first OpenClaw session with Critiqor. Four commands: install, monitor, run, finalize — then your diagnosis dashboard opens automatically.

Critiqor wraps your OpenClaw session from start to finish: it creates a run, attaches its evidence plugin, launches OpenClaw, silently observes everything that happens, and — when you’re done — generates a diagnosis and opens a local dashboard in your browser. The entire workflow runs through four commands and requires no manual configuration.

1. Install Critiqor

   Install Critiqor from PyPI:

   pip install critiqor
   

   This installs the critiqor CLI and the bundled OpenClaw plugin. No separate plugin setup is required.

2. Start monitoring

   Start a monitored OpenClaw session:

   critiqor monitor openclaw
   

   Critiqor runs through the following automatically before OpenClaw ever starts:

   • Generates a unique run_id (e.g. run_001) and creates a runs/run_001/ directory
   • Writes an active_session.json file to track the in-progress session
   • Attaches the bundled OpenClaw evidence plugin
   • Initializes the runtime observer and begins capturing events
   • Launches openclaw chat in the same terminal as a supervised child process

   Because the observer starts before openclaw chat launches, every event from the very beginning of the session is captured.

   OpenClaw must be installed and available on your PATH for this command to work. Critiqor launches it via openclaw chat internally.

3. Use OpenClaw normally

   The OpenClaw TUI opens in your terminal. Use it exactly as you normally would — send prompts, invoke tools, and interact with your agent. You do not need to change anything about how you work.

   While you interact with OpenClaw, Critiqor silently observes and captures:

   • Tool calls and tool outputs
   • Memory events
   • Retries and errors
   • Context and state transitions
   • Token usage
   • The full runtime event timeline

   When you’re finished, exit the OpenClaw TUI as you normally would (e.g. Ctrl+C or the built-in exit command).

4. Finalize the session

   After exiting OpenClaw, finalize the session to generate your diagnosis and open the dashboard:

   critiqor finalize
   

   Critiqor will:

   1. Close the active session and stop the runtime observer
   2. Read the raw session.json evidence collected during the run
   3. Run the diagnosis engine — detecting failure modes, building the causal graph, calculating the trust score, and generating recommendations
   4. Write diagnosis.json alongside session.json in the run directory
   5. Start a local dashboard server on an available port
   6. Open the dashboard in your browser automatically at http://127.0.0.1:<port>/?run_id=run_001

   You’ll see output similar to the following in your terminal:

   Stopping observer...
   Finalizing evidence...
   Generating diagnosis...
   Diagnosis saved: runs/run_001/diagnosis.json
   Starting local dashboard...
   Dashboard run: run_001
   Critiqor dashboard: http://127.0.0.1:<available-port>/?run_id=run_001
   

   Critiqor validates diagnosis.json before launching the dashboard. If the diagnosis is missing or invalid, dashboard launch is aborted and the CLI prints a clear recovery message instead of opening stale or default data.

Files Created

After finalizing, your run artifacts are organized as follows:

runs/
  run_001/
    session.json    # raw runtime evidence (tool calls, events, timeline)
    diagnosis.json  # diagnosis, trust score, causal graph, recommendations
active_session.json # tracks the in-progress session (removed after finalize)

session.json is the complete structured runtime session — timeline events, tool activity, memory events, runtime metadata, and aggregate metrics. diagnosis.json is derived from that evidence and contains the trust score, primary failure mode, causal chain, severity, impact scores, and recommended fixes. The original evidence in session.json remains auditable and separate.

What to Do Next

Once the dashboard opens, you’ll see an Overview page with your trust score and a readiness verdict. From there you can:

• Check the trust score — a 0–100 reliability score weighted by the evidence Critiqor captured
• Read the diagnosis — the primary failure mode detected, its causal chain, and the recommended fix
• Explore the Evidence view — drill into the raw tool calls, outputs, memory events, retries, and the full execution trace
• Review the causal graph — a step-by-step explanation of why the session failed (if a failure was detected)

To revisit a past session or open a specific run’s dashboard later, use:

# List all completed runs with summaries
critiqor runs

# Reopen the dashboard for a specific run
critiqor dashboard run_001

For a full reference of all CLI commands and their options, see the CLI Reference. To learn more about reading and navigating the dashboard, see the Dashboard Guide.