The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code diffs for security analysis. While this is the intended purpose, it creates a surface for indirect prompt injection where an attacker might attempt to bypass the LLM-based logic review (Step 3) using obfuscated or maliciously crafted diffs. The skill explicitly mitigates this by enforcing chunking and using deterministic SAST tools.
Ingestion points: Code diffs processed by the LLM in the Step 3: Logic & PII Check section.
Boundary markers: The guardrails section explicitly mentions enforcing strict chunking or rejecting diffs that exceed token limits to prevent context manipulation.
Capability inventory: The skill utilizes subprocess calls to execute security tools like gitleaks, semgrep, npm audit, and pip-audit.
Sanitization: No explicit string sanitization is mentioned, but the skill relies on qualifying findings with confidence scores.
[COMMAND_EXECUTION]: The skill invokes industry-standard CLI tools for security auditing. These commands (gitleaks, semgrep, npm audit, pip-audit) are used for their primary defensive purpose and do not represent unauthorized or malicious command execution.

safety-scan