The Agent Skills Directory

[PROMPT_INJECTION]: No malicious instructions, behavior overrides, or safety bypass attempts were detected in the skill body or instructions.
[DATA_EXFILTRATION]: The skill does not perform network operations, hardcode credentials, or access sensitive system files such as SSH keys or environment secrets.
[REMOTE_CODE_EXECUTION]: No external scripts are downloaded or executed. The tool relies on a local Python script for static analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code, which constitutes a potential surface for indirect prompt injection.
Ingestion points: The scripts/scan_deserialization.py script reads local files through Path.read_text().
Boundary markers: The scanner report includes extracted code snippets wrapped in backticks but lacks explicit instructions for the agent to ignore any commands contained within those snippets.
Capability inventory: The skill uses Read and Bash(python3:*) tools to perform its function.
Sanitization: Extracted code snippets are truncated to 160 characters and stripped of leading/trailing whitespace. This surface is fundamental to the skill's purpose as a scanner and is assessed as safe in this context.

detecting-insecure-deserialization