The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes standard security auditing tools via the shell, including 'npm audit', 'pip audit', 'cargo audit', and 'git log'. These operations are performed to identify vulnerabilities and exposed secrets within the target project's environment.
[CREDENTIALS_UNSAFE]: The skill defines various regex patterns to detect sensitive information like AWS keys, Stripe tokens, and private keys. To mitigate exposure risks, the skill mandates a masking protocol (showing only the first and last four characters) for all reported secrets.
[PROMPT_INJECTION]: The skill processes untrusted external data (source code and project configuration files) during its audit phase, creating a surface for indirect prompt injection.
Ingestion points: Project source files and configuration files (e.g., 'package.json', '.wardstones/config.json').
Boundary markers: Not explicitly defined in the instructions to isolate or delimit untrusted data.
Capability inventory: Execution of shell commands for auditing and writing report files to the project directory.
Sanitization: The skill includes context-aware filtering logic for secrets detection but lacks explicit sanitization or delimiters for potential natural language instructions embedded in the analyzed code.

heimdall