The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data such as application logs, source code, and configuration files during the audit workflow. This creates a potential surface for indirect prompt injection if the audited data contains instructions designed to manipulate the agent.
Ingestion points: Local files and log traces read via Bash (ripgrep) and Read tools.
Boundary markers: The instructions do not specify explicit delimiters or "ignore" markers for the processed data.
Capability inventory: Read, Write, Edit, and Bash tools are available to the agent.
Sanitization: No explicit sanitization or filtering of the audited content is described.
[SAFE]: The skill uses standard local analysis tools for its primary diagnostic purpose. No attempts to exfiltrate data, bypass safety guardrails, or execute remote code from untrusted sources were identified. The use of bash commands is restricted to searching local files.

agent-architecture-audit