critiqor check — Run Agent Deployment Policy Checks

critiqor check validates your evaluation history against configurable thresholds. Use it to gate deployments on trust score and hallucination risk limits.

critiqor check reads your saved evaluations and validates them against reliability thresholds that you define. Use it to gate deployments, enforce reliability standards in CI, or verify that your agent meets a required quality bar before shipping.

Syntax

critiqor check [OPTIONS]

What It Does

The command reads your evaluation file and checks whether the results meet your specified thresholds. It exits 0 if all checks pass and a non-zero code if any threshold is not met. You can define thresholds directly as flags or supply a policy file for reusable, version-controlled configuration.

Options

Flag	Type	Default	Description
--evaluations	path	critiqor_evaluations.jsonl	Path to the evaluation file to check.
--agent-id	string	(none)	Filter checks to a specific agent.
--policy	path	(none)	Path to a JSON or YAML policy file defining thresholds.
--minimum-trust-score	integer	(none)	Minimum trust score (0–100) required to pass.
--maximum-hallucination-risk	integer	(none)	Maximum hallucination risk (0–100) allowed.

Examples

# Run with explicit thresholds
critiqor check --minimum-trust-score 75 --maximum-hallucination-risk 20

# Filter by agent
critiqor check --agent-id my_agent --minimum-trust-score 80

# Use a policy file
critiqor check --policy policy.json

Example policy.json

{
  "minimum_trust_score": 75,
  "maximum_hallucination_risk": 25
}

Exit Codes

Code	Meaning
0	All checks passed — every threshold was met.
Non-zero	One or more thresholds were not met.

CI Integration

Because critiqor check uses standard exit codes, it works in any shell-based CI pipeline. Add it as a step after your evaluation workflow to block a deployment when reliability drops below your threshold.

Run critiqor finalize --no-dashboard first to ensure your evaluation file is up to date before the check runs:

critiqor finalize --no-dashboard && critiqor check --policy policy.json

Make sure you have finalized your most recent session with critiqor finalize before running policy checks. Unfinalized sessions are not included in the evaluation file and will not be reflected in the check results.