The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/download-stitch-asset.sh performs network requests using Python's urllib.request to download assets from URLs provided at runtime. While intended for fetching screenshots from Google Cloud Storage, the script does not validate the target domain.
[COMMAND_EXECUTION]: The skill utilizes a shell script scripts/download-stitch-asset.sh to manage asset downloads and references a feedback recording script utilities/skill-builder/scripts/record_skill_feedback.py to be executed via python3 as part of the SKILL.md workflow.
[EXTERNAL_DOWNLOADS]: The README.md file provides installation instructions using npx skills add google-labs-code/stitch-skills, which involves downloading skill components from a remote source associated with a well-known technology organization.
[PROMPT_INJECTION]: The skill ingests untrusted metadata such as screen titles and descriptions from the external Stitch platform. This content is used for video overlays and composition generation. Ingestion points: screens.json and external Stitch project data. Boundary markers: None identified. Capability inventory: Subprocess execution via shell scripts and Remotion CLI. Sanitization: No explicit validation or escaping of external text inputs is documented, creating a surface for indirect prompt injection.

stitch-remotion