surgical-diff
Skill1 starsChanged 7 months agoMITView on GitHub
Review a code diff for scope creep, drive-by changes, and style drift. Use when the user pastes a diff, asks "review this diff", "did I change too much", "is this change surgical", or wants to audit an AI-generated patch before merging. Returns flagged lines, scope analysis, a minimal-diff proposal, and a verdict.
In this repository11 other files it gives its agents.
- assumption-surfaceSkill
- description-craftSkill
- goal-criteriaSkill
- karpathy-codingSkill
- lean-responseSkill
- simplicity-auditSkill
- skill-compressSkill
- skill-graphSkill
- token-auditSkill
- token-budget-checkSkill
- vixero-councilSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 61
After the edit, are imports / variables / functions orphaned?
- Distinctive: line 108
Never recommend deleting pre-existing dead code unless user asked.
- Distinctive: line 110
If stated intent is missing, do not analyze.
- Distinctive: line 111
If the diff is under 5 lines, skip and say "too small to drift".
- Distinctive: line 115
Never "fix" the diff yourself.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
surgical-diff
Audit a diff against its stated intent. Every changed line must trace to the request.
Trigger on
- "review this diff"
- "is this surgical"
- "did I change too much"
- "audit this patch"
- "scope-check this PR"
- User pastes a diff plus a statement of what the change was supposed to do
Input shapes
| shape | handling |
| unified diff + stated intent | full analysis |
| diff only, no intent | ask for intent first, do not guess |
| file-before + file-after | treat as implicit diff |
| PR description + diff | extract intent from description |
| commit message + diff | extract intent from subject line |
Analysis passes
Pass 1 — scope trace
For each changed line: does this line solve the stated request? Mark yes / no / indirect.
Pass 2 — drift detection
Universal drift flags:
| drift type | examples |
| style | quote style changes, whitespace, trailing commas |
| type theatre | type hints added to unrelated functions |
| doc theatre | docstrings added to unrelated functions |
| rename | variable / function renames not required by fix |
| scope creep | new features adjacent to the fix |
| unauthorized delete | pre-existing dead code removed |
| dependency creep | new import not needed by the fix |
| test theatre | unrelated tests added alongside the fix |
Language-specific drift:
| language | flag when you see |
| Python | f-string conversion of unrelated % or .format() lines; list → List type annotation backport on unrelated functions |
| TypeScript | any → specific type on unrelated signatures; interface → type or reverse |
| JavaScript | var → const/let on unrelated lines; function → arrow on unrelated |
| Go | error wrap reformat on unrelated errors; interface{} → any swap on unrelated |
| Rust | unwrap() → ? conversion on unrelated call sites |
| Java | var local-type inference backport on unrelated locals |
Pass 3 — orphan check
After the edit, are imports / variables / functions orphaned?
| orphan type | action |
| Import your edit orphaned | remove |
| Variable your edit orphaned | remove |
| Function your edit orphaned | remove |
| Orphan that pre-existed | leave. flag only if user asks about it |
Output
stated intent: {one line}
lines changed: {n}
lines tracing to intent: {n} ({pct}%)
drift flags:
style: {line numbers}
type-theatre: {line numbers}
doc-theatre: {line numbers}
rename: {line numbers}
scope-creep: {line numbers + description}
unauthorized-delete: {line numbers}
dependency-creep: {line numbers}
language-specific: {type + line numbers}
orphans-created: {list — remove}
orphans-untouched: {list — leave, informational}
minimal diff (only lines tracing to intent):
```diff
{the subset}verdict: {surgical | minor drift | scope creep | rewrite needed} reasoning: {one line}
## Verdict thresholds
| tracing % | verdict |
|-----------|---------|
| ≥ 95 | surgical |
| 80–94 | minor drift (acceptable with flags) |
| 50–79 | scope creep (split into multiple PRs) |
| < 50 | rewrite needed (start from stated intent again) |
## Hard rules
- Never recommend deleting pre-existing dead code unless user asked.
- Style drift is a flag, not an error. Flag it, let the user decide.
- If stated intent is missing, do not analyze. Ask for it first.
- If the diff is under 5 lines, skip and say "too small to drift".
- Type hints on NEW code do not flag. Type hints added to UNRELATED code do.
- Tests added that cover the stated fix do NOT flag. Tests unrelated to the fix do.
- Whitespace-only diffs (reformat commits) are flagged as their own category — suggest splitting into a separate PR.
- Never "fix" the diff yourself. Report and propose. The user decides.Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

