agent-report
Warn
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill clones the
agent-report-kitrepository from GitHub (https://github.com/fang2hou/agent-report-kit) and executes its content usingpnpmandnodeto run a local interactive service on the host machine. - [COMMAND_EXECUTION]: The skill performs multiple shell operations including directory creation, package installation via
pnpm, running JavaScript services vianode, capturing headless screenshots using Chromium-based binaries (Chrome, Chromium, or Edge), and opening generated files in the system's default browser. - [EXTERNAL_DOWNLOADS]: The skill fetches a renderer toolkit from a GitHub repository (
https://github.com/fang2hou/agent-report-kit) as part of its initialization workflow. - [INDIRECT_PROMPT_INJECTION]: The skill processes external user input through a local state file (
state.json) produced by the interactive report service, creating a surface for potential instruction injection. - Ingestion points: Reads
state.jsonwhich contains user decisions, answers, and chat events from the local web interface. - Boundary markers: None identified in the instructions for delimiting the user-provided data when read by the agent.
- Capability inventory: File system writes, subprocess execution (
pnpm,node, Chromium), and local network service hosting. - Sanitization: Relies on the external toolkit's schema validation during the JSON build process.
Audit Metadata