The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by design, as its primary purpose is extracting and processing data from external sources.
Ingestion points: The skill ingests data using extract_excel_files, extract_from_pdfs, and extract_from_api functions (SKILL.md), which handle content from files and network responses.
Boundary markers: There are no explicit boundary markers or system instructions provided to the agent to isolate and ignore potentially malicious natural language instructions embedded within the processed Excel sheets, PDF tables, or JSON API payloads.
Capability inventory: The skill is granted filesystem and network permissions (claw.json) and includes functional code for file reading/writing, database interaction via sqlite3, and issuing network requests using the requests library (SKILL.md).
Sanitization: Although the code includes data quality functions like clean_construction_data and validate_construction_data (SKILL.md), these focus on structural and numeric integrity rather than filtering or sanitizing text for adversarial instructions.

etl-pipeline