critiqor monitor openclaw — Start Agent Observation

critiqor monitor openclaw launches OpenClaw with runtime observation. Options: —agent-id, —runs-dir, —difficulty-tier, —timeout, —visibility.

critiqor monitor openclaw launches the OpenClaw TUI via openclaw chat and wraps it with Critiqor’s runtime observer. While the session is active, Critiqor captures tool calls, state transitions, process events, and plugin evidence — everything needed to produce an accurate diagnosis when you later run critiqor finalize.

Syntax

critiqor monitor openclaw [OPTIONS] [-- AGENT_COMMAND...]

Arguments

Argument	Description
AGENT_COMMAND (optional)	Legacy: pass -- <command> [args...] to run and observe a custom agent command instead of launching OpenClaw interactively.

Options

Option	Default	Description
--agent-id TEXT	openclaw_agent	Agent identifier used in run metadata and dashboard
--tenant-id TEXT	default	Tenant identifier for multi-agent setups
--visibility	private	Evidence visibility: private, public, anonymous, or shared
--events TEXT	.critiqor/events.jsonl	Legacy event log path for custom command runs
--evaluation TEXT	.critiqor/latest_run.json	Output path for the latest run diagnosis JSON
--benchmark-id TEXT	openclaw_runtime_v1	Versioned benchmark spec ID
--difficulty-tier	standard	Benchmark difficulty tier: easy, standard, hard, or stress
--runs-dir TEXT	runs	Directory where run artifacts are stored
--openclaw-command TEXT	openclaw chat	Command used to launch OpenClaw
--cwd TEXT	(current directory)	Working directory for the OpenClaw process
--timeout FLOAT	(none)	Optional timeout in seconds — terminates OpenClaw if exceeded

What happens when you run it

1. Checks for an existing active session — if runs/active_session.json already exists, the command exits with instructions to run critiqor finalize first.
2. Verifies openclaw is on PATH — resolves the executable from --openclaw-command. If not found, exits with an error and installation guidance.
3. Creates a new session in runs/<run_id>/ with status MONITORING and writes runs/active_session.json.
4. Sets environment variables so the Critiqor plugin inside OpenClaw knows where to write evidence: CRITIQOR_RUN_ID, CRITIQOR_RUNS_DIR, CRITIQOR_AGENT_ID, CRITIQOR_TENANT_ID, CRITIQOR_VISIBILITY, and CRITIQOR_EVENT_SOURCE.
5. Launches openclaw chat as a child process with the plugin active, printing confirmation before handing control to the TUI:
   ✓ OpenClaw detected
   ✓ Runtime observer attached
   ✓ Event collection active
   
   Launching OpenClaw...
   
6. Waits for the OpenClaw session to end, forwarding the process exit code.

After OpenClaw exits

When the OpenClaw process terminates (whether by user exit, error, or --timeout), Critiqor prints:

OpenClaw session ended.
Run `critiqor finalize` to stop monitoring and generate a diagnosis report.

The collected evidence remains in runs/<run_id>/ until you finalize. Nothing is lost if you close your terminal — run critiqor finalize from the same working directory to complete the session.

Troubleshooting

“OpenClaw command not found” : OpenClaw is not installed or is not on your PATH. Install OpenClaw, or pass --openclaw-command with the full path to the executable, for example:

critiqor monitor openclaw --openclaw-command /usr/local/bin/openclaw chat

“Critiqor monitoring is already active for <run_id>” : A previous session was never finalized. Run critiqor finalize to close it before starting a new one.