Auditor

valdr-auditor is a thin wrapper around auditor prompts. It does not hardcode the scoring rubric itself. It resolves which auditor should handle the session, loads that auditor’s prompt, and then defers the audit method to the loaded prompt.

Auditor is available on Vanguard and Sovereign plans with MCP access.

Use Auditor When You Need To

• Audit a Valdr session
• Evaluate how well an agent executed a task
• Record an audit score run with a specific auditor persona
• Compare audit behavior across different auditor prompts

How To Use It

Start with:

/valdr-auditor

Then point it at the session or task you want scored:

• Basic audit prompt:

Audit the latest executor session for WEB-142.

• Named auditor prompt:

Use tyr-v2 to audit the latest executor session for WEB-142 and record a score run.

• Session-specific prompt:

Use tyr-v2 to audit session 01JX8Y6J9J3M6P4T4Y6R8ABCDE and summarize the biggest execution risks.

If you omit the auditor handle, the wrapper will list the available auditors and ask you which one should own the audit.

Wrapper Contract

The wrapper always performs three steps:

1. Resolve the auditor handle.
2. Load the selected auditor prompt.
3. Follow the loaded prompt as the single source of truth.

Step 1: Resolve the Auditor

If the user already names an auditor handle, use it directly.

If the user does not name one, list the available auditor agents:

mcp__valdr__pm_agent → { "action": "list", "defaultRoles": ["auditor"] }

Resolution rules:

• If one or more auditors exist, present the handles and ask the user which auditor should own the audit.
• If none exist, stop and route the user to Orchestrator for registry repair or setup.

Step 2: Load the Prompt

mcp__valdr__pm_agent → { "action": "get_prompt", "handle": "<auditor handle>" }

If the prompt cannot be loaded, stop and ask the user how to proceed.

Step 3: Execute the Loaded Audit Workflow

Once the prompt is loaded, that prompt owns the evidence model, score dimensions, severity bands, persistence flow, and any follow-up references it hot-loads.

Why the Wrapper Stays Thin

Auditing standards change. Different auditors may use different rubrics. Keeping those rules in the loaded prompt lets you update the audit behavior centrally while the wrapper remains stable and easy to reason about.

Related Skills

• Reviewer — Review the work itself before or alongside process evaluation
• Executor — Produce the task sessions that an auditor later evaluates
• Orchestrator — Route registry setup or broader workspace discovery before auditing