The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The discovery workflow described in reference/discovery.md instructs the agent to ingest content from potentially untrusted local files like .cursorrules, AGENTS.md, and CLAUDE.md, as well as to tour live websites. These ingestion points lack explicit boundary markers or sanitization, potentially allowing malicious instructions embedded in the project or site to influence subsequent file-writing or script-execution actions.
[DYNAMIC_EXECUTION]: The skill uses @remotion/bundler in scripts/stills.ts and scripts/render.ts to bundle React source code into executable assets for rendering. It also uses dynamic imports for local configuration files like webpack-override.ts, which is a standard procedure within the Remotion ecosystem.
[EXTERNAL_DOWNLOADS]: The skill utilizes uv and npx to fetch and execute dependencies such as remotion, imageio-ffmpeg, and numpy from official package registries (NPM and PyPI). These are well-known services and the packages are standard for this domain.
[COMMAND_EXECUTION]: Multiple scripts including scripts/audio-edit.py, scripts/beats.py, scripts/render.ts, and scripts/verify.py execute shell commands to interface with ffmpeg and the Remotion CLI. These operations are performed using safe invocation methods (e.g., argument lists) that minimize the risk of shell injection.

product-film