The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains references to injection-marker strings such as 'ignore previous instructions'. These occur exclusively within documentation and tables describing how to detect prompt injection attacks in LLM applications. No attempt to bypass safety guidelines or override agent instructions was found.\n- [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze potentially untrusted data, including application logs and source code. This represents an indirect prompt injection surface; however, the skill is designed to identify and categorize such risks as part of its auditing function.\n- [DYNAMIC_EXECUTION]: Instructions include the use of shell commands like ripgrep and xargs to automate the discovery of security-relevant code patterns. These are standard diagnostic techniques appropriate for the skill's stated purpose.\n- [COMMAND_EXECUTION]: The 'purple-team replay' section provides curl commands intended to simulate attack traffic (e.g., SQLi probes or brute-force attempts) against a staging environment to verify detection capabilities.

aegis-blue-team-detection