Reviewer

valdr-reviewer is a routing wrapper around reviewer prompts. Its job is to fetch the task first, decide which reviewer handle should own the review, load that reviewer’s prompt, and then defer the review procedure to the loaded prompt.

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

Use Reviewer When You Need To

• Run a review for an existing task
• Continue a task that already has reviewer assignments
• Start a fresh review with a specific reviewer handle
• Route a task into the verification gate with the right reviewer prompt loaded

How To Use It

Start with:

/valdr-reviewer

Then ask for the review you want:

• Basic review prompt:

Review task WEB-142.

• Named reviewer prompt:

Review task WEB-142 with reviewer sigrid.

• New assignment with extra focus:

Start a review for task OPS-31 with reviewer security-reviewer and focus on risky shell commands and permission changes.

If the task already has reviewer assignments, the wrapper will use those assignments as the source of truth before it loads the reviewer prompt.

Wrapper Contract

The wrapper follows a three-part sequence:

1. Fetch the task.
2. Resolve or start the right review assignment.
3. Load the reviewer prompt for that handle and follow it.

Step 1: Fetch the Task

mcp__valdr__pm_task → { "action": "get", "taskKey": "<task-key>" }

The task response determines whether reviewer assignments already exist.

Step 2: Resolve the Reviewer

If the task already has reviewer assignments:

• Present the assigned reviewer handles to the user.
• If the user named a handle that is not assigned, stop with an error.

If the task has no reviewer assignments:

• If the user already named a reviewer, start a review with that handle.
• If the user did not name one, list available reviewers and ask the user to choose.

To list available reviewer agents:

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

To create a new review assignment when needed:

mcp__valdr__pm_review → { "action": "start", "taskKey": "<task-key>", "reviewerHandle": "<reviewer handle>" }

Step 3: Load the Reviewer Prompt

Once the reviewer handle is resolved, load the prompt:

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

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

What Lives Outside the Wrapper

Severity rules, score interpretation, verification-gate policy, and comment style all live in the loaded reviewer prompt or any references it hot-loads.

That means the wrapper stays stable while the actual review standard can evolve.

Related Skills

• Executor — How agents execute work before review
• Auditor — Evaluate the quality of the resulting session after review
• Planner — Create the structured work that eventually lands in review
• Orchestrator — Route registry setup or broader workspace discovery before review starts