The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables an agent to browse and record arbitrary external websites. This create a risk where instructions embedded on a malicious website could attempt to override agent behavior during the recording process.
Ingestion points: Multiple scripts in SKILL.md use page.goto() to load untrusted content from the web.
Boundary markers: Absent. The generated scripts do not include specific delimiters or instructions to ignore text found on the target pages.
Capability inventory: The skill demonstrates file system access (fs.renameSync), command execution (execSync in reference.md), and browser-based network interaction.
Sanitization: Absent. The skill does not filter or sanitize the content of the pages it visits.
[COMMAND_EXECUTION]: In reference.md, the skill provides a utility function using execSync to run the ffprobe command.
Evidence: execSync(ffprobe -v error -show_entries format=duration -of csv=p=0 "${filePath}"). While this is a standard use case for analyzing recorded video files, it involves launching a sub-process with variables in the command string.
[DYNAMIC_EXECUTION]: The skill frequently uses page.evaluate() to inject and execute JavaScript in the browser context.
Evidence: Multiple examples in SKILL.md inject CSS and JS for cursor visualization and scroll effects. This is a standard and necessary functionality for the skill's purpose (UI recording).

playwright-recording