The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a human-in-the-loop workflow where the agent ingests untrusted text, highlights, and flags from an external web studio to modify source files.
Ingestion points: Data is captured from a locally served web studio and returned to the agent via a structured JSON patch or URL query parameters, as described in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when processing the ingested human feedback.
Capability inventory: The agent is instructed to perform file-write operations ('applies the inline edits to the source file', 'bakes the chosen state into reproducible assets') and execute local servers to host the studio.
Sanitization: There are no instructions for sanitizing or validating the human-provided text before it is baked back into the production code or assets.
[DYNAMIC_EXECUTION]: The skill requires the agent to generate and execute code at runtime to provide the interactive tuning environment.
Runtime generation: The agent is instructed to create a temporary, single-page HTML/JS studio in a scratch directory based on the artifact being tuned.
Dynamic script creation: The workflow requires the agent to 'port the renderer to Python / build script / etc.' at runtime to 'bake' the human-selected state into final assets, which involves generating executable logic based on the studio's state output.

lookdev