Module review 

bn review — Adversarial post-close review of bean implementations.

After an agent closes a bean, a review agent checks the implementation against the bean’s spec in a fresh context, providing semantic correctness checking beyond what verify gates can catch.

Flow

1. Load bean description + acceptance criteria
2. Collect git diff (changes since HEAD)
3. Build a review prompt with spec + diff + verdict instructions
4. Spawn review agent (using config.review.run or config.run template)
5. Parse VERDICT from agent output: approve / request-changes / flag
6. Apply verdict: update labels, optionally reopen bean with notes

Verdicts

• approve — implementation correct; adds reviewed label
• request-changes — issues found; reopens bean with notes, adds review-failed
• flag — needs human attention; adds needs-human-review label, stays closed

Structs

ReviewArgs

Arguments for cmd_review.

Enums

ReviewVerdict

Verdict returned (or inferred) from the review agent’s output.

Functions

apply_verdict

Apply the parsed verdict to the bean: update labels, notes, and status.

cmd_review

Execute bn review <id>.

parse_verdict

Parse the review agent’s output for a VERDICT keyword.