The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes two Node.js scripts, capture.mjs and rebrand.mjs, which perform subprocess operations. Specifically, capture.mjs executes Playwright (Chromium) and rebrand.mjs uses spawnSync to execute project-local "hook" scripts.
[DYNAMIC_EXECUTION]: The rebrand.mjs script dynamically executes a user-provided script path (the "hook") defined in a JSON mapping file. Additionally, capture.mjs performs dynamic loading of the playwright module using createRequire and import() from the project's local directory.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external content from project briefs and design tool outputs to generate prompts, standing headers, and review notes. This represents a potential surface for indirect injection attacks.
Ingestion points: Reads content from docs/brief.md and processes human-provided pastes of design tool outputs.
Boundary markers: The skill relies on natural language instructions and template placeholders (e.g., <project>) but does not implement explicit security delimiters or "ignore embedded instructions" warnings for ingested content.
Capability inventory: The skill has capabilities to write to the file system (e.g., creating the prototypes/ directory and updating CLAUDE.md/AGENTS.md) and execute scripts (capture.mjs, rebrand.mjs).
Sanitization: No specific sanitization or validation logic is defined for the external text interpolated into the generated prompts.
[EXTERNAL_DOWNLOADS]: The capture.mjs script facilitates the use of the Playwright library, which is a well-known service from Microsoft. It includes logic to handle the download of the Chromium browser via the standard npx playwright install chromium command if it is not already present in the environment.

pin-prototypes