The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.check_output across several reference files to interact with local system binaries.
Evidence: Calls to git (in mine-change-coupling.md, mine-bus-factor.md, mine-hotspots-churn-complexity.md, mine-bug-fix-density.md, local-history-prior-localization.md, mine-codebase-aging.md, mine-commit-topic-modeling.md, sim-doc-code-alignment.md).
Evidence: Calls to gh (GitHub CLI) in sim-cross-pr-feature-mapping.md to fetch pull request metadata.
Evidence: Calls to gumtree in clone-ast-gumtree.md for AST differencing.
[EXTERNAL_DOWNLOADS]: The instructions and code examples recommend the installation and use of a wide array of third-party Python packages for specialized analysis.
Evidence: gensim (topic modeling), networkx (graph analysis), scikit-learn (VSM/TF-IDF), transformers (CodeBERT/embeddings), spacy (NLP), nltk (Porter stemming), pydivsufsort (suffix arrays), complexipy (cognitive complexity), radon (cyclomatic complexity), sentence_transformers (embeddings), datasketch (MinHash/LSH), simhash (fingerprinting), apted (tree edit distance), rapidfuzz (string matching), rank-bm25 (IR ranking).
[DATA_EXFILTRATION]: The mine-bus-factor algorithm processes personally identifiable information (PII) by extracting developer email addresses from the local version control system.
Evidence: references/mine-bus-factor.md uses git blame --line-porcelain to collect author-mail fields for authorship concentration metrics.
[SAFE]: No evidence of prompt injection, code obfuscation, or persistence mechanisms was found. The skill is highly technical and provides accurate references to canonical academic papers and established software engineering metrics.

linguistic-semantic-algorithms