The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing the flue package from PyPI (pip install flue). This package is authored by the vendor sfkislev and is necessary for the skill's functionality. The instructions explicitly state that the agent should not install the package without human approval.
[COMMAND_EXECUTION]: The skill interacts with desktop applications by executing local bridge scripts, such as figma_bridge.py, and using the flue CLI for setup and pairing.
[DYNAMIC_EXECUTION]: The bridge facilitates automation by sending scripts (e.g., JavaScript for Figma or ExtendScript for Adobe apps) to the target application's internal scripting runtime. The skill also mentions the ability to execute raw JavaScript using an eval command through the bridge.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external applications which could contain malicious instructions.
Ingestion points: Untrusted data enters the agent context via structured JSON returned by figma_bridge.py (referenced in SKILL.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used for the returned JSON data.
Capability inventory: The skill can execute shell commands via figma_bridge.py and flue, and execute dynamic JavaScript within applications (SKILL.md).
Sanitization: No explicit sanitization or filtering of the JSON output is described, although the skill emphasizes human oversight and small steps.

Figma