The Agent Skills Directory

[PROMPT_INJECTION]: The LinkedIn content analysis feature in scripts/signal_scanner.py is vulnerable to indirect prompt injection.
Ingestion points: The skill ingests untrusted text from LinkedIn posts via the Apify linkedin-profile-posts actor.
Boundary markers: The LLM prompt template in _llm_score_post directly interpolates post content without using delimiters (e.g., XML tags) or providing specific instructions to the LLM to ignore embedded commands.
Capability inventory: The LLM's output directly controls logic that writes to the signals database table and updates the lead_status of records in the people table.
Sanitization: No input validation, filtering, or escaping is performed on the external text before it is processed by the AI model.

signal-scanner