agent-governance-gap-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONOBFUSCATION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a bundled Python script (
scripts/scan.py) to automate the discovery and inspection of local AI agent configurations and environment variables.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data found in local configuration files (e.g.,mcp.json,.env,settings.json), creating a surface for indirect prompt injection.\n - Ingestion points: The skill reads and processes local filesystem configuration and environment files via the
scripts/scan.pyscript.\n - Boundary markers: The instructions direct the agent to present findings as a table and seek explicit user confirmation before scoring, reducing the risk of autonomous obedience to injected instructions.\n
- Capability inventory: The skill utilizes read-only filesystem access and local script execution; it does not possess network or file-write capabilities.\n
- Sanitization: The scanning script implements a robust redaction layer (
sanitize_url,redact_text) that filters out credential values, connection string passwords, and high-entropy tokens before findings reach the agent context.\n- [OBFUSCATION]: The scanning script and its documentation use minor string fragmentation (e.g.,github_pat+_) in regex patterns to represent sensitive token prefixes, intended to prevent the skill's own source code from being flagged by static analysis tools.
Audit Metadata