product-feedback-report
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Node.js script
report-workspace.mjsto manage the feedback lifecycle and spawns a headless Chrome/Chromium process to generate PDF reports. The commands use hardcoded arguments and array-based spawning, which mitigates shell injection risks. - [DATA_EXFILTRATION]: The
report-workspace.mjsscript allows copying arbitrary files into the report workspace as 'evidence'. While designed for screenshots, the lack of source path validation means the agent could be directed to copy sensitive local files (e.g., SSH keys or configuration files) into the report directory if misled. This risk is partially mitigated by instructions for the agent to inspect and redact evidence. - [PROMPT_INJECTION]: The skill processes untrusted user-supplied feedback which is rendered into an HTML and PDF report. This creates a surface for indirect prompt injection.
- Ingestion points: Findings (titles, happened, expected, impact) provided by the user and processed by the
addandupdatecommands inscripts/report-workspace.mjs. - Boundary markers: Data is passed via structured JSON through
stdin, separating agent instructions from data. - Capability inventory: Filesystem access to write reports and config, and subprocess execution to run Chrome.
- Sanitization: The script uses
escapeHtmlandescapeCssStringto sanitize finding content before rendering, reducing the risk of injected scripts executing in a browser or PDF viewer.
Audit Metadata