scrape-codegen-analyze

Pass

Audited by Gen Agent Trust Hub on Jul 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted HTML data to generate analysis for a downstream agent. Malicious instructions hidden in HTML elements, JSON-LD, or scripts could influence the agent's behavior.
  • Ingestion points: Raw HTML file (page_html_path) and metadata.
  • Boundary markers: None. Untrusted content is processed without delimiters or 'ignore' instructions.
  • Capability inventory: Access to Bash (shell execution), Read, and Write tools.
  • Sanitization: None. The skill explicitly preserves scripts during cleaning, which may contain malicious payloads.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (mkdir, uv run) using variables derived from the $ARGUMENTS string. This creates a surface for command injection if the arguments (such as work_path) contain shell metacharacters.
  • [DATA_EXPOSURE]: The skill accesses files outside its immediate directory (e.g., ../scrape/references/python-environments.md), which relies on the agent having broad read permissions across the workspace.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 18, 2026, 09:08 AM
Security Audit — agent-trust-hub — scrape-codegen-analyze