The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by processing untrusted external data.
Ingestion points: The skill ingests data from external websites via WebFetch and audits local repository source code.
Boundary markers: The instructions lack explicit delimiters or safety warnings to ignore embedded instructions within the ingested web content or code files (e.g., HTML comments or JS comments).
Capability inventory: Across its workflow, the skill employs Bash, Write, and Edit tools, which provide the ability to execute arbitrary shell commands and modify local files.
Sanitization: No validation or sanitization of the external data is performed before it is analyzed by the agent, creating a risk if that data contains malicious instructions targeting the agent's capabilities.
[REMOTE_CODE_EXECUTION]: The skill utilizes dynamic execution of JavaScript code to perform its core functions.
Evidence: The references/browser-profiling.md file contains several large JavaScript templates designed to be executed via tab.playwright.evaluate within a browser tab context. While intended for performance profiling, this pattern involves executing dynamically provided code at runtime.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform repository management and project builds.
Evidence: It explicitly instructs the agent to run git status, npm run build, and npm run lint. It also describes a pattern for isolated git commits using a temporary GIT_INDEX_FILE in /tmp/.

optimize-web-animations