The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions contain no behavioral overrides, jailbreak attempts, or safety filter bypasses. Instructional language is consistent with the stated purpose of performing a security audit.
[DATA_EXFILTRATION]: While the skill intentionally searches for sensitive strings like api_key and secret, this is for the purpose of generating a local security report. There is no evidence of network operations or attempts to transmit found secrets to external domains.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface as it is designed to ingest and analyze untrusted codebase content which could contain malicious instructions.
Ingestion points: Project source directories (e.g., src/, assets/data/) and configuration files.
Boundary markers: None explicitly defined in the instructions for the sub-agent task.
Capability inventory: Bash, Write, Task, Grep, Read.
Sanitization: No specific sanitization or filtering of external content before processing is mentioned.
[COMMAND_EXECUTION]: The skill uses Bash and Grep for local file searching. These operations are restricted to the codebase context and do not involve suspicious command patterns or privilege escalation.

security-audit