| name | gf-change-audit |
| description | Split large GF Framework changes into focused review tracks before commit, release, or risky integration. Use after broad refactors, API changes, layer moves, extension restructuring, reference-project migration work, or when the user asks for multi-agent audit. |
GF Change Audit
Use this skill to review large GF changes without mixing concerns. When sub-agents are available, run independent read-only tracks in parallel; otherwise run the same tracks sequentially.
Prepare
- Pin the review scope:
- For committed work, ask or infer a base and use
git diff <base>...HEAD.
- For dirty work, inspect
git status --short --branch, git diff --name-status, and staged diff if present.
- List touched files and classify them with
python tools\gf_maintenance.py workspace-status --json.
- Identify the user's task intent, acceptance criteria, and explicit out-of-scope boundaries. If no spec exists, audit only against repository rules and stated conversation intent.
Parallel Tracks
Use the prompts and output format in review-axes.md. Keep each track bounded and independent:
- Standards:
AI_MAINTENANCE.md, CODING_STYLE.md, API_SURFACE.md, layer boundaries, generated-doc rules, release rules.
- API and compatibility: public API surface,
@since, schema docs, SemVer, deprecated/removed behavior, generated reference drift.
- Verification: tests, docs checks, examples checks, release checks, skipped commands, and whether failures were expected.
- Reference boundary: external
gf-reference-project, sync behavior, smoke paths, framework feedback, and project-business leakage.
- Architecture: module depth, locality, dependency direction, extension atomicity, and whether a new seam earns its interface cost.
Aggregate
Report blocking findings first, then warnings, then clean tracks. Cite local file paths and commands for concrete review findings. Do not merge away disagreement between tracks; preserve it as an explicit open question.
When a track finds a deterministic rule that is not automated yet, recommend whether it belongs in tests/gf_core/maintenance, tools/gf_maintenance.py, or AI_MAINTENANCE.md.
