The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of bookmark titles and URLs during the audit and classification phases.
Ingestion points: The workflow in SKILL.md (Step 2: Audit Before Designing) reads the complete bookmark tree into the agent's context.
Boundary markers: The skill uses a structured assets/dry-run-plan-template.json to organize data and separate it from operational instructions.
Capability inventory: The skill possesses the ability to move, rename, and delete bookmarks using browser automation tools.
Sanitization: The risk is mitigated by a mandatory dry-run plan that requires human review before any bulk or destructive operations are executed.
[DYNAMIC_EXECUTION]: The skill utilizes browser automation and the execution of scripts within the context of internal browser pages to manage favorites.
Evidence: references/chrome-edge-execution.md specifies the use of Playwright and CDP (Chrome DevTools Protocol) to interact with chrome://bookmarks and edge://favorites. It incorporates safety patterns such as pilot runs and fresh state verification to manage risks associated with automated interaction.
[COMMAND_EXECUTION]: The skill instructs the agent to use automation tools to send input and commands to browser-internal management interfaces.
Evidence: Specifically targets automation of the bookmark manager UI and API surface, utilizing focus and keyboard delivery as detailed in the execution reference files.

browser-bookmark-organizer