scratch-html
Pass
Audited by Gen Agent Trust Hub on Jun 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local shell script,
index.sh, to regenerate theindex.htmldashboard whenever a new file is created. - [EXTERNAL_DOWNLOADS]: The
index.shscript references typography and style assets from Google Fonts (fonts.googleapis.comandfonts.gstatic.com) to render the generated index dashboard. - [PROMPT_INJECTION]: Potential surface for indirect prompt injection or data poisoning within the generated
index.htmldashboard. - Ingestion points: The
index.shscript parses the<title>metadata from all HTML files located in the~/.scratch/directory to build the navigation list. - Boundary markers: Absent. Titles extracted from external files are interpolated into the dashboard without explicit delimiters or warnings for the agent.
- Capability inventory: The skill allows the agent to read and write files within the user's home directory (specifically
~/.scratch/) and execute local bash scripts. - Sanitization: The script performs minimal sanitization on extracted titles using a basic regular expression (
sed -e 's/<[^>]*>//g'), which may be insufficient if a processed file contains malicious instructions or scripts designed to influence the agent or browser.
Audit Metadata