The Agent Skills Directory

[PROMPT_INJECTION]: Documentation in references/audit-output-and-workflow.md and detection logic in scripts/prompt_injection_scanner.py contain known prompt injection strings like 'ignore previous instructions'. These are identified as signatures for the scanner to flag in external files and do not influence the agent's core instructions.
[REMOTE_CODE_EXECUTION]: References to dangerous functions like eval(), exec(), and compile() appear in scripts/code_scanner.py and references/threat-model-and-patterns.md. These are part of the tool's signature database for identifying risks in target code and are not called by the skill's own logic.
[OBFUSCATION]: The skill documents common obfuscation techniques, including base64 and hex encoding, in its threat model. It scans for these patterns in target files using scripts/code_scanner.py. These are descriptive detection patterns, not malicious payloads.
[COMMAND_EXECUTION]: The tool identifies command injection vectors such as os.system and subprocess with shell=True. The scanner's own implementation uses safe argument handling via argparse and does not execute untrusted commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (markdown and code files) to generate audit reports. While this provides a theoretical ingestion point for malicious content, the risk is minimized as the tool uses static regex-based parsing and the agent is instructed to treat the output as a diagnostic report.
Ingestion points: scripts/code_scanner.py, scripts/prompt_injection_scanner.py, and scripts/supply_chain_checker.py read content from target file paths.
Boundary markers: The scripts produce structured text reports intended for terminal display or JSON output.
Capability inventory: The Python scripts utilize standard libraries for file I/O and regex matching; they do not possess autonomous execution capabilities over the scanned content.
Sanitization: Content is processed via static pattern matching (regex) rather than dynamic evaluation.

skill-security-auditor