The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze untrusted data sources (source code, request bodies, and configuration files), which provides a surface for indirect prompt injection if those files contain malicious instructions disguised as comments or data.\n- Ingestion points: The agent is directed to ingest target codebases, API payloads, and configuration files (YAML, XML, JSON) as described in the 'What To Check' and 'Where To Look' sections of SKILL.md.\n- Boundary markers: There are no instructions for the agent to use specific delimiters or 'ignore' markers when processing the audited content to isolate instructions from data.\n- Capability inventory: The skill implies access to file-reading tools and shell execution (for vulnerability reproduction steps), which increases the potential impact of a successful injection.\n- Sanitization: The instructions do not provide methods for the agent to sanitize or filter external content before it is processed in the agent's context.

aegis-deserialization