The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python scripts (scripts/setup_asset_pipeline.py and scripts/remove_chroma_key.py) that use the subprocess.run module to execute shell commands. These commands manage virtual environments, install packages, and call image processing tools like magick or identify. Additionally, references/research-method.md instructs the agent to use ffmpeg for temporal decomposition of video assets.
[EXTERNAL_DOWNLOADS]: The setup_asset_pipeline.py script installs external software to the local environment. It uses pip or uv to install the Pillow library and utilizes system package managers (such as brew, apt-get, dnf, pacman, or winget) to install the ImageMagick suite.
[PRIVILEGE_ESCALATION]: The setup script for ImageMagick includes commands that utilize sudo for Linux-based package managers (e.g., sudo apt-get install). The documentation in references/asset-acquisition.md specifies that the agent must require explicit user approval before performing these administrator-level actions.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves researching and ingesting external data from design galleries, URLs, and code snippets, which presents a surface for indirect prompt injection.
Ingestion points: According to references/research-method.md, the skill ingests briefs, screenshots, videos, URLs, live sites, and code fragments.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded prompts within the ingested research material.
Capability inventory: The skill possesses the ability to execute shell commands via subprocess, perform network operations via Browser or MCP tools, and write files to the project directory.
Sanitization: There is no evidence of sanitization or filtering applied to external content before it is processed by the agent.

balise-webmotion