The Agent Skills Directory

[OBFUSCATION]: The execution script scripts/main.cjs utilizes hex escapes to encode the string 'eval' as 'e\x76al' within its vulnerability detection regexes.\n
Evidence: Multiple patterns in scripts/main.cjs such as pattern: /e\\x76al\\s*\\([^)]*(?:req|user|input|param)/gi use this encoding method.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code from local directories and outputs snippets in its report, creating a surface for indirect prompt injection.\n
Ingestion points: Files are read from the target directory using fs.readFileSync in scripts/main.cjs.\n
Boundary markers: No delimiters or 'ignore' instructions are used when presenting extracted code snippets to the agent.\n
Capability inventory: The skill employs the Bash, Read, Write, and WebFetch tools.\n
Sanitization: Extracted snippets are included in the output without validation, escaping, or filtering for malicious instructions.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill makes network requests to the well-known OSV.dev service to retrieve dependency vulnerability data.\n
Evidence: Uses WebFetch to query https://api.osv.dev/v1/querybatch as documented in SKILL.md and scripts/main.cjs.

gemini-cli-security