review-results

name	review-results
description	Review existing git-slop report and health artifacts, explain detector and overlay evidence, and optionally turn one explicitly selected finding into a bounded maintenance proposal. Use when a user asks what results mean, which findings are actionable, or wants a plan from a reviewed finding; do not use merely to generate a fresh report.

Review Git Slop Results

Start from an existing .slop/latest/report.json, report.yaml, summary.md, or health.md artifact.

Interpret The Evidence

1. Treat git-slop as a detector and evidence surface, not a correctness oracle.
2. Keep hotspot costs separate from overlay evidence. Use slop_score, slop_band, and context_band for the stable hotspot queue.
3. Use health file/folder bands, distributions, watchlists, and concentration metrics as human-facing rollups of existing facts, not as a second score.
4. Treat health output as advisory: findings do not change the command's success status or the stable check gate.
5. Use overlays only as supporting evidence about organization, verification, navigation, blast radius, stewardship, and concept dispersion.
6. Follow a finding's deterministic next_command, or run a targeted git-slop explain, before recommending work.
7. Treat budget_exceeded as a threshold label and review candidate, not an automatic mandate to change code.
8. Give each reviewed finding one explicit disposition: implement, consolidate, defer, accept, or won't fix. Do not optimize for closing every finding.
9. Stop after explaining and prioritizing the evidence unless the user explicitly asks for a maintenance proposal.

Plan One Reviewed Finding

When the user explicitly selects a finding for planning, read the maintenance-planning reference and follow it for that finding only.

If the proposal would add or promote a command, option, configuration key, schema, output, Action contract, Agent Plugin surface, workflow, integration, or compatibility promise, also read the contract-admission reference. Apply its state model, admission test, and surface-area ledger before recommending implementation.

Preserve Generated-State Boundaries

Treat .slop/latest/, .slop/runs/, .slop/cache/, SARIF, prompt packs, plan JSON, and compare JSON as generated evidence. Reference or upload a bounded artifact when useful; do not commit routine generated output.