The Agent Skills Directory

[SAFE]: The skill implements a security enforcement layer with a strict 'deny-by-default' policy. It utilizes the python-magic library for robust file type identification and references local domain blacklists (e.g., Sci-Hub) to prevent access to known high-risk sources. No credentials, data exfiltration, or obfuscation were detected.
[COMMAND_EXECUTION]: The scripts/run_security_scan.py file is a local runner script that initializes an AgentSession to execute the 'security-scan' stage. It includes logic to handle command-line arguments and writes scanning reports to a specific project directory (examples/snse_survey/skill_outputs). This behavior is transparent and aligned with the skill's stated purpose as a developer tool or system component.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it processes external target files and URLs provided during runtime.
Ingestion points: The target input field in SKILL.md (local file paths or external URLs).
Boundary markers: None explicitly defined in the prompt instructions to isolate content from the agent's logic.
Capability inventory: The script invokes AgentSession.run_stage, which triggers the internal JinYiWeiAgent for scanning operations.
Sanitization: The skill relies on file type detection and domain matching rather than content sanitization, representing a standard surface for indirect injection if the processed content is later interpreted by an LLM without guardrails. Given its role as a security scanner, this is a known operational surface rather than a malicious finding.

security-scan