The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted external data from LinkedIn feeds and comments.
Ingestion points: The agent reads data from the LinkedIn feed (https://www.linkedin.com/feed/), user activity pages (https://www.linkedin.com/in/{username}/recent-activity/all/), and specific post comment threads as described in SKILL.md (Workflow 1 & 2).
Boundary markers: Absent. The instructions do not provide delimiters or specific 'ignore embedded instructions' warnings for the data extracted from LinkedIn.
Capability inventory: The skill leverages powerful browser automation tools (claude-in-chrome and chrome-devtools-mcp) allowing for keyboard/mouse interaction, JavaScript execution, and navigation. It also includes local file system capabilities for drafting content in .md files.
Sanitization: Absent. Content extracted via JavaScript or snapshots (e.g., post text, comments) is interpolated into the agent's context without sanitization or validation.
[COMMAND_EXECUTION]: The skill includes predefined JavaScript snippets in SKILL.md and references/profile-config.md for data extraction (e.g., fetching post metadata and profile URLs). These are intended for use with the browser automation tools and are static rather than dynamically generated from untrusted input.

linkedin-chrome