The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a "clipboard loader" pattern in SKILL.md that uses eval() to execute arbitrary JavaScript code retrieved from the system clipboard via navigator.clipboard.readText(). This mechanism allows for the execution of unverified logic that bypasses static security analysis.
[COMMAND_EXECUTION]: The provided automation scripts, such as 1c-screenshot.js and scan-nav.js, utilize Node.js fs and path modules when executed through the browser_run_code tool. This grants the skill the ability to create directories, write selector cache files, and delete old files on the host system filesystem, specifically within the playwright-screenshots/ and .claude/ directories.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests untrusted data from the web application's interface. Scripts like 1c-snapshot.js scrape labels and values from forms, and scan-nav.js extracts navigation titles and IDs. This data is used to inform the agent's next steps without explicit sanitization or instructions to ignore embedded commands within the application UI.

1c-web-session