The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external APIs and local files, creating a potential surface for indirect prompt injection attacks.
Ingestion points: The skill fetches data from the Apollo Company and People Search APIs and the SixtyFour enrichment service in scripts/tam_builder.py. It also reads data from local JSON files in scripts/recover_sixtyfour_people.py.
Boundary markers: No explicit sanitization or instruction-isolating delimiters are implemented in the Python scripts to prevent the interpretation of malicious instructions that might be embedded in company keywords, bios, or other external metadata.
Capability inventory: The skill has significant capabilities, including performing outbound network requests to multiple third-party APIs and executing write, patch, and upsert operations on Supabase database tables (companies and people).
Sanitization: While the scripts perform basic string normalization such as trimming and lowercasing, they do not include filters to detect or neutralize potential prompt injection content that could influence the AI agent's subsequent analysis of the database results.

tam-builder