The Agent Skills Directory

COMMAND_EXECUTION (HIGH): The script scripts/detect-secrets.sh is vulnerable to shell command injection via the target directory path.
Evidence: The script uses eval to execute a find command where the $TARGET_DIR variable is interpolated without sanitization or escaping: eval "find '$TARGET_DIR' -type f ...". A maliciously crafted directory name (e.g., '; touch /tmp/pwned; #') provided as the target would cause the eval call to execute arbitrary shell commands.
File: scripts/detect-secrets.sh
PROMPT_INJECTION (LOW): The skill is susceptible to indirect prompt injection as it processes untrusted data with powerful capabilities.
Ingestion points: Source code and configuration files within the $TARGET_DIR scanned by the full-audit.sh and detect-secrets.sh scripts.
Boundary markers: Absent. The scripts do not utilize delimiters or explicit instructions to prevent the agent from interpreting malicious instructions found within the scanned codebase.
Capability inventory: The skill possesses shell execution capabilities (via Bash for npm audit, pip-audit, find, and grep) and performs extensive directory traversal and file reading.
Sanitization: Absent. Input variables like $TARGET_DIR and the contents of scanned files are processed without validation or sanitization.
File: scripts/full-audit.sh, scripts/detect-secrets.sh
Metadata Poisoning (MEDIUM): The skill documentation is misleading regarding its functional completeness.
Description: SKILL.md and full-audit.sh reference scripts/owasp-check.py and scripts/generate-report.py as core components for OWASP compliance auditing and report generation. These files are not present in the package, meaning the skill cannot perform its primary advertised security functions.

security-auditor