Install any skill in seconds. Free to start, no credit card required.
Get Started Free →This skill should be used when the user asks to "fix a bug with minimal changes", "analyze change scope for a bugfix", "find the minimal set of files to change", "do a focused bugfix", or "scope a minimal repair".
.claude/skills/borghei-focused-fix/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 32% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 697% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 48% | 0% |
> Category: Engineering > Domain: Debugging & Maintenance
The Focused Fix skill enforces a disciplined minimal-change approach to bug fixing. Instead of refactoring or improving code during a bugfix, it identifies the smallest possible change set that resolves the issue. This reduces risk, simplifies code review, and prevents scope creep.
Before scoping the fix, confirm these inputs. If any is unknown or vague, ASK — do not assume:
--path/--extensions)Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.
bash# Analyze a bug description to identify minimal change scope python scripts/change_scope_analyzer.py --bug "Login fails when email has + character" --path ./src # Analyze with JSON output python scripts/change_scope_analyzer.py --bug "API returns 500 on empty array input" --path ./src --format json # Analyze with specific file extensions python scripts/change_scope_analyzer.py --bug "CSS overflow on mobile" --path ./src --extensions .css .scss .html
| Tool | Purpose | Key Flags | |------|---------|-----------| | change_scope_analyzer.py | Identify minimal files to change for a bugfix | --bug, --path, --extensions, --format |
Analyzes a bug description against a codebase to identify:
change_scope_analyzer.py to identify scopegit diff --stat matches (or is smaller than) the analyzer's recommendationgit diff shows no formatting-only changes, no unrelated imports, no "while I'm here" edits| Anti-pattern | Failure mode | Fix | |--------------|--------------|-----| | "While I'm here" refactors in the bugfix PR | Blast radius explodes; review time multiplies; unrelated regressions masked by the real fix | Open a separate PR for the refactor, tagged as refactor: not fix: | | Reformatting or auto-save style changes in touched files | Diff becomes unreadable; real fix hidden in 200 lines of whitespace | Revert style changes before committing; configure the editor to format-on-save only for new files | | Fixing the symptom in the wrong layer | Bug returns in a new form; accumulates workaround debt | Trace to the root layer — analyzer's keyword-match output is a hint, not an answer | | Skipping the regression test "because the fix is obvious" | Bug silently returns on a refactor 6 months later | Every fix: commit adds at least one failing-then-passing test | | Treating change_scope_analyzer.py output as authoritative | Analyzer is keyword/import-based, not semantic — misses dynamic dispatch, reflection, config-driven paths | Use it as a starting set; grep for callers and tests before committing to the scope | | Bundling the fix with a dependency upgrade | Two risk profiles in one PR; if rollback is needed, both are lost | Land the upgrade separately, then the fix against the upgraded baseline |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 11,005 | 5,092 | -54% | 1 | 1 | 0% | 1,588 | 2,099 | +32% | 0 | 0 | — |
case-01 | fail→pass | 2,792 | 15,081 | +440% | 1 | 1 | 0% | 400 | 3,186 | +697% | 0 | 0 | — |
case-02 | fail→fail | 13,153 | 12,637 | -4% | 1 | 1 | 0% | 2,182 | 3,343 | +53% | 0 | 0 | — |
case-03 | fail→fail | 12,523 | 22,047 | +76% | 1 | 1 | 0% | 2,058 | 4,432 | +115% | 0 | 0 | — |
case-04 | pass→pass | 8,149 | 1,744 | -79% | 1 | 1 | 0% | 1,218 | 1,655 | +36% | 0 | 0 | — |
case-06 | pass→pass | 10,304 | 7,188 | -30% | 1 | 1 | 0% | 1,528 | 2,356 | +54% | 0 | 0 | — |
case-07 | pass→pass | 10,904 | 9,564 | -12% | 1 | 1 | 0% | 1,682 | 2,853 | +70% | 0 | 0 | — |
case-08 | fail→pass | 10,991 | 5,934 | -46% | 1 | 1 | 0% | 1,612 | 2,202 | +37% | 0 | 0 | — |
case-09 | pass→pass | 10,243 | 7,050 | -31% | 1 | 1 | 0% | 1,575 | 2,424 | +54% | 0 | 0 | — |
case-10 | pass→fail | 12,503 | 10,747 | -14% | 1 | 1 | 0% | 1,839 | 2,937 | +60% | 0 | 0 | — |
case-11 | fail→pass | 12,445 | 9,273 | -25% | 1 | 1 | 0% | 1,880 | 2,683 | +43% | 0 | 0 | — |
case-12 | fail→pass | 11,334 | 7,969 | -30% | 1 | 1 | 0% | 1,718 | 2,540 | +48% | 0 | 0 | — |
case-13 | fail→pass | 10,049 | 1,918 | -81% | 1 | 1 | 0% | 1,654 | 1,595 | -4% | 0 | 0 | — |
case-14 | pass→pass | 12,524 | 8,500 | -32% | 1 | 1 | 0% | 1,934 | 2,599 | +34% | 0 | 0 | — |
case-15 | pass→pass | 13,637 | 8,424 | -38% | 1 | 1 | 0% | 1,858 | 2,511 | +35% | 0 | 0 | — |
case-16 | pass→pass | 9,053 | 4,911 | -46% | 1 | 1 | 0% | 1,314 | 2,027 | +54% | 0 | 0 | — |
case-17 | pass→pass | 12,851 | 5,757 | -55% | 1 | 1 | 0% | 1,934 | 2,263 | +17% | 0 | 0 | — |
case-18 | fail→pass | 8,392 | 2,317 | -72% | 1 | 1 | 0% | 1,346 | 1,691 | +26% | 0 | 0 | — |
case-19 | pass→pass | 11,889 | 8,085 | -32% | 1 | 1 | 0% | 1,810 | 2,482 | +37% | 0 | 0 | — |
case-20 | pass→pass | 9,288 | 7,189 | -23% | 1 | 1 | 0% | 1,374 | 2,411 | +75% | 0 | 0 | — |
case-21 | pass→fail | 27,763 | 10,650 | -62% | 1 | 1 | 0% | 4,796 | 2,934 | -39% | 0 | 0 | — |
case-22 | pass→pass | 28,017 | 20,180 | -28% | 1 | 1 | 0% | 4,814 | 4,932 | +2% | 0 | 0 | — |
case-23 | pass→pass | 10,834 | 10,222 | -6% | 1 | 1 | 0% | 1,551 | 2,980 | +92% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +22 percentage points is the difference between those two pass rates over the 23 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.