The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/lint_design_md.py uses subprocess.run to execute the official Google design linter command npx -y @google/design.md lint. This execution is controlled and targets a specific tool from a trusted organization for design validation. The script scripts/test_design_system.py also uses subprocess.run to execute Python scripts within the skill directory for self-testing purposes.
[EXTERNAL_DOWNLOADS]: The skill performs network operations using urllib.request. scripts/find_fonts.py fetches font metadata from Google's official fonts metadata endpoint. scripts/scan_site.py fetches HTML and CSS from external URLs provided by the user for competitive design analysis. These downloads are central to the skill's functionality and are performed in a read-only manner.
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface as scripts/scan_site.py ingests data from external, untrusted websites. 1. Ingestion points: scripts/scan_site.py reads content from user-specified URLs. 2. Boundary markers: The skill instructions emphasize separating observed data from inferred judgments and documenting limitations in the Evidence section of the generated documentation. 3. Capability inventory: The skill can execute shell commands via npx (for linting) and write local design specification files. 4. Sanitization: The skill uses html.escape when generating the HTML specimen preview in scripts/build_specimen.py to prevent script injection in the generated output.

fth-design-crafter