scrape
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from untrusted external web pages, which serves as a vector for malicious instructions.
- Ingestion points: Web content is retrieved via browser snapshots, text extraction, and agent-based research tools.
- Boundary markers: The skill contains a dedicated 'Untrusted content' warning block, instructing the agent to treat page results as content only and never as instructions.
- Capability inventory: The agent possesses
Bashexecution,Readtool access, andAskUserQuestioncapabilities. - Sanitization: Explicit rules prohibit the execution of commands, code, or tool calls found within page content.
- [COMMAND_EXECUTION]: The skill invokes several vendor-specific binaries located in the user's home directory for setup, telemetry, and logging purposes.
- Evidence: Execution of
~/.claude/skills/gstack/bin/gstack-skill-start,gstack-skill-end, andgstack-learnings-logduring the skill's lifecycle. - [DYNAMIC_EXECUTION]: The skill generates JavaScript code at runtime to be executed within the browser's context via the
aside replcommand. - Evidence: JavaScript templates in the scraping steps are used to define selectors and extraction logic dynamically based on the user's intent.
Audit Metadata