The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted codebases and configurations, creating a potential surface for indirect prompt injection from the materials being audited. 1. Ingestion points: Code and configuration files are accessed through the read_file and search_code tools. 2. Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions to encapsulate audited data. 3. Capability inventory: The skill utilizes powerful tools such as execute_command, run_build, and run_tests to perform technical assessments. 4. Sanitization: The instructions focus on redacting secrets from final reports but do not provide specific mechanisms for sanitizing untrusted input before processing.
[SAFE]: The skill instructions promote security best practices, such as the redaction of secrets, failure-closed error handling, and adherence to established security frameworks. No evidence of malicious intent, obfuscation, or unauthorized exfiltration was found.

security-audit