The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from job descriptions and resumes to inform the profile rewriting process, creating a vulnerability surface where adversarial text could influence the agent's behavior.
Ingestion points: External text files for job descriptions (e.g., hunter/jd/) and resumes (e.g., hunter/resumes/) are read by the profile.py script and the agent.
Boundary markers: None identified. The instructions do not specify delimiters or guidelines for the agent to differentiate between user instructions and potentially adversarial content within the job descriptions.
Capability inventory: Analysis of profile.py shows no subprocess calls, usage of eval() or exec(), file-writing operations, or network requests. The script strictly performs read-only text processing and keyword matching using standard libraries.
Sanitization: The skill does not sanitize external content. However, profile.py includes a confirm function that programmatically flags terms in the draft that do not appear in the source resume or current profile, acting as a partial mitigation against injected or hallucinated data.

hunter-linkedin