The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and analyze untrusted game source files (like GameScene.js) to determine how to patch game logic and generate input sequences. This creates a surface where malicious code comments or identifiers in the source could influence the agent's script generation.
Ingestion points: Analyzes local project files (e.g., GameScene.js) to find method names for collision and game-over logic.
Boundary markers: None. The skill does not instruct the agent to use delimiters or ignore embedded instructions when reading source code.
Capability inventory: The skill generates a JavaScript execution file (scripts/capture-promo.mjs) and executes shell commands including npx playwright and ffmpeg.
Sanitization: The skill lacks sanitization logic for the patterns extracted from the analyzed source code before including them in the generated script template.
[DYNAMIC_EXECUTION]: The skill uses a template-based approach to generate a functional Node.js script (scripts/capture-promo.mjs) at runtime, which is then executed to control the browser.
Evidence: The SKILL.md file provides a 'Full Capture Script Template' and instructions for the subagent to adapt and write this file to the project directory before execution.
[COMMAND_EXECUTION]: The skill performs multiple shell command executions to facilitate video recording and post-processing.
Evidence: The skill executes npx playwright --version, ffmpeg -version, and a bundled shell script scripts/convert-highfps.sh to perform video encoding and FPS adjustments.

promo-video