movp-review

MoVP Adversarial Review

Scope: plan files and significant code output
MCP tools: get_review_status, trigger_review, resolve_review

After writing a plan file or producing a significant code change (new feature, refactor, migration, or multi-file edit), trigger an adversarial review using the MoVP tools. Do not trigger for trivial operations: file reads, git status/log, search queries, single-line fixes, or tool calls with no code output.

When to trigger

Trigger trigger_review automatically when you:

• Write or significantly modify a plan file (.md planning documents, architecture docs, PRDs)
• Complete a multi-file code change (3+ files, or any new service/module)
• Produce a significant new code output in a single response (functions, handlers, migrations)

Do not trigger for: reading files, running tests, git commands, grep/search, single-line edits, config-only changes.

How to run a review

1. Call trigger_review(artifact_type="plan_file"|"code_output", content=<artifact>, session_id=<current session>)
   → returns review_id

2. Poll get_review_status(review_id=<id>) until review_status is "completed" or "error"
   Always use the review_id from step 1 — do NOT call get_review_status without review_id
   when multiple reviews may be in flight (returns most recent tenant review otherwise).

3. Ask the developer: implement fixes, dismiss findings, or accept as-is

4. Call resolve_review(review_id=<id>, action="accept"|"dismiss"|"escalate"|"retry") based on their choice

Presenting findings

Format findings as structured output with severity badges. After showing findings, always ask:

Reply with: implement fixes, dismiss (false positive / not applicable / deferred), or accept as-is

Resolve actions

Developer says	Action to call	Notes
"accept", "looks good", "ship it"	resolve_review(action="accept")	Idempotent — safe to call twice
"dismiss", "false positive", "not applicable"	resolve_review(action="dismiss", reason="false_positive"|"not_applicable"|"deferred")
"escalate", "create a ticket"	resolve_review(action="escalate", target="todo")
"retry", "run it again"	resolve_review(action="retry")	Only valid when review_status is "error" — do not call on completed reviews

Rate and cost awareness

Reviews consume LLM budget. Do not trigger multiple reviews in a single session for the same artifact. If trigger_review returns a rate limit error (429), inform the developer and do not retry automatically.