The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from articles stored in the local repository to check against technical gates. This creates a surface for indirect prompt injection where malicious instructions embedded in a reviewed post could attempt to override the agent's audit logic or hijack its tools. \n
Ingestion points: Technical articles located in authority-blog/content or learn/content, and corpus index files such as learn/plan/CONTENT_INDEX.md. \n
Boundary markers: The skill lacks explicit structural delimiters to isolate the article content from the agent's instructions, though it internally directs the agent to treat content as passive evidence via its Security Guardrails section. \n
Capability inventory: The agent has access to read_file, write_file, edit_file, create_file, and search_code tools, which could be abused if an injection is successful. \n
Sanitization: No explicit sanitization, validation, or escaping of the article content is specified before it is processed by the agent.

audit-technical-article