1---
 2name: review-packet
 3kind: service
 4---
 5
 6# Review Packet
 7
 8### Description
 9
10Write the final Crosscheck review for a human or other LLM agent (likely not a
11Prose VM) who may act now or file issues.
12
13### Requires
14
15- `brief`: review brief
16- `runner_capabilities`: runner capability report
17- `lane_plan`: active and skipped lane plan
18- `lane_reports`: reviewer lane reports
19- `candidate_ledger`: closed candidate ledger
20- `validation_report`: validator verdicts
21- `merged_findings`: merged finding groups
22
23### Ensures
24
25- `review_packet`: final semi-structured Markdown review packet
26
27### Shape
28
29- `self`: put findings first, then human-judgment items, rejected/downgraded
30  candidates, coverage, limits, and next steps
31- `prohibited`: presenting degraded coverage as complete, hiding skipped lanes,
32  hiding safety caveats, inventing fixes, or promoting follow-up concerns to
33  findings
34
35### Strategies
36
37- Use stable headings: Summary, Findings, Needs human judgment, Rejected or
38  downgraded candidates, Coverage, Limitations, Suggested next steps.
39- Before formatting `review_packet`, read `references/output-format.md`.
40- Keep prose readable in chat while preserving candidate IDs and source lanes.
41- Always report the resolved reviewed workspace.
42- Always report eligible lane count, active/skipped lanes, skipped runner
43  reasons, providers/models use, families and harnesses represented, validator
44  degradation, and any safety violation.