The Agent Skills Directory

[DATA_EXFILTRATION]: The security design review script iterates through and reads the contents of every file in the targeted directory. This behavior potentially exposes sensitive information such as private keys, configuration secrets, or local data if the tool is run against a directory containing them. * Evidence: scripts/security_design_review.py performs a recursive search using Path(path).rglob('*') and reads each file's content into memory for keyword matching.
[CREDENTIALS_UNSAFE]: The threat modeling script explicitly identifies and targets sensitive environment configuration files as assets for analysis. * Evidence: scripts/threat_model.py includes .env in its list of config_files to search for and identify as project assets.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and performs static analysis on untrusted code and configuration files from the local filesystem, creating a surface for indirect prompt injection. * Ingestion points: Files are read in scripts/analyze_patterns.py, scripts/identify_spof.py, and scripts/security_design_review.py. * Boundary markers: The scripts do not implement boundary markers or instructions for the agent to ignore content within the analyzed files. * Capability inventory: The skill has read access to the local filesystem but lacks network egress or file-write capabilities. * Sanitization: No input sanitization or content filtering is performed on the files prior to analysis.

architect-reviewer