The Agent Skills Directory

[METADATA_POISONING]: The skill name, description, and internal script documentation consistently claim the tool performs Natural Language Processing tasks such as sentiment analysis and keyword extraction. However, the provided script scripts/analyze_text.py contains no NLP logic and instead performs file system metadata analysis, including directory scanning and file size reporting. This discrepancy is deceptive and could lead an agent to perform broad file system scans while the user expects a text analysis operation.\n- [INDIRECT_PROMPT_INJECTION]: The skill's core purpose is to ingest and process external text data, which is an untrusted input source. Without specific instructions to isolate this data, it represents an attack surface for instructions embedded in analyzed text to influence the agent.\n
Ingestion points: User-provided text files and content processed via 'analyze', 'review', and 'examine' triggers defined in SKILL.md.\n
Boundary markers: Absent; no delimiters or instructions to ignore embedded commands are included in the skill's prompting.\n
Capability inventory: Broad permissions including Bash(cmd:*), Read, Grep, and Glob as listed in the YAML frontmatter.\n
Sanitization: Absent; the skill defines no methods for validating, filtering, or escaping content extracted from the analyzed documents.

analyzing-text-with-nlp