The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits external web pages using browser automation, which introduces a potential surface for indirect prompt injection. \n- Ingestion points: The scripts in references/browser-profiling.md extract textContent, animation names, and class names from the DOM of audited websites using Playwright evaluation. \n- Boundary markers: The skill lacks specific instructions or markers to distinguish between data and instructions when processing extracted text content. \n- Capability inventory: The skill has the ability to modify local source code, perform git commit operations, and execute npm build commands. \n- Sanitization: While scraped text is truncated to 80 characters, there is no evidence of filtering or escaping for potential malicious instruction sequences. \n- [COMMAND_EXECUTION]: The skill utilizes standard command-line tools for repository management and verification. \n- Evidence: SKILL.md instructs the agent to execute git status, npm run build, and perform atomic commits using temporary index files in the /tmp directory. \n- [DYNAMIC_EXECUTION]: The skill uses dynamic loading for environment-specific utilities. \n- Evidence: references/browser-profiling.md utilizes an import() statement with a placeholder path to load the browser client module at runtime.

optimize-web-animations