SKILL.md
Find Dead Code
Detect unused code that can be safely removed.
Rules
- Report findings, do not delete. Removal is a separate task.
- Focus on production code (
ols/). Skip tests unless explicitly asked. - Vulture has false positives — classify each finding before recommending removal.
- Code used only via dynamic dispatch (e.g. Pydantic validators, FastAPI dependencies) is not dead.
Step 1: Determine Scope
Ask the user:
- Branch mode: only files changed in the current branch vs main.
- Full mode: scan the entire
ols/directory.
For branch mode:
git diff --name-only origin/main -- 'ols/' | grep '\.py$'Step 2: Run Vulture
uvx vulture <target> --min-confidence 80--min-confidence 80 reduces noise. Lower confidence findings are more likely false positives.
Step 3: Run Pylint Unused Checks
uv run pylint --disable=all --enable=unused-import,unused-variable,unused-argument,unreachable <target>Cross-reference with vulture findings to increase confidence.
Step 4: Filter False Positives
Common false positives in this codebase:
| Pattern | Why it's not dead |
|---|---|
Pydantic modelvalidator, fieldvalidator |
Called by Pydantic, not directly |
| FastAPI dependency functions | Injected via Depends() |
eq, hash, str |
Called implicitly by Python |
| Constants used in config YAML | Referenced by config loader |
| Abstract method implementations | Called via base class interface |
Imports re-exported from init.py |
Used by external consumers |
Step 5: Classify Findings
For each finding, classify:
| Category | Criteria | Action |
|---|---|---|
| Remove | Clearly unused, no dynamic references | Safe to delete |
| Verify | Possibly used dynamically or externally | Search for string references before removing |
| False positive | Pydantic/FastAPI/magic method | Skip |
For "Verify" findings, search for string-based references:
rg "<function_or_class_name>" ols/ tests/Step 6: Report
For each finding:
- File and line number
- What is unused (function, class, import, variable)
- Confidence level (vulture %)
- Classification (remove / verify / false positive)
Summary: total findings, how many safe to remove, estimated lines saved.