@alexvegman/cli

The agentguard command-line tool for AgentGuard — captures, reviews, and repairs AI agent tool-use behavior.

Install

npm install -g @alexvegman/cli

Commands

| Command | What it does | |---|---| | agentguard init | Scaffold an agentguard.config.ts in the current project. | | agentguard watch | Capture a live agent run (from a transcript file or a wrapped command) and evaluate it. | | agentguard test | Run AgentGuard's assertion suite against a captured run. | | agentguard review-pr | Review a git diff for correctness/test/secret hygiene — used in CI. | | agentguard replay | Re-run the evaluation pipeline over a stored run with no browser, agent, or network. | | agentguard report | Render a stored run's results (console, JSON, JUnit, HTML). | | agentguard compare | Diff two runs' results. | | agentguard calibrate | Check assertion calibration/confidence against @alexvegman/mutations profiles. | | agentguard audit-fixtures | Verify golden/correct-behavior fixtures are actually discriminable. | | agentguard autofix propose / show | Propose and inspect an automatic fix from a pattern of failing runs. | | agentguard export-pack | Export a run directory into a self-contained, integrity-checked evidence pack. | | agentguard doctor | Diagnose local environment/config issues. | | agentguard history | List and inspect prior runs. | | agentguard tokens | Report token usage for a run. | | agentguard --version | Print the installed version and build commit. |

Depends on

@alexvegman/core, @alexvegman/decision, @alexvegman/assertions, @alexvegman/reporters, @alexvegman/mutations.

Full documentation: github.com/AlexanderVegazo26/agent-guard.