The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local system commands to perform its reformatting and output tasks. * Evidence: The skill uses pbpaste to ingest content and pbcopy to output the cleaned results to the clipboard. * Evidence: The skill uses the open command to launch browsers for viewing generated HTML files. * Evidence: Mentions the possible use of pandoc for document format conversion.
[DATA_EXFILTRATION]: The skill accesses system data and internal configuration paths. * Evidence: It reads the full content of the system clipboard, which often contains sensitive user data or temporary credentials. * Evidence: It references a specific internal agent memory file located at ~/.claude/memory/feedback_social_link_placement.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the clipboard or user prompts without explicit boundary markers, creating a surface for injection. * Ingestion points: Data is pulled directly from the clipboard via pbpaste or from user-provided prompt text. * Boundary markers: The instructions lack specific delimiters or directions for the agent to ignore instructions embedded within the text being cleaned. * Capability inventory: The skill has the power to write to the clipboard (pbcopy), write files to the /tmp directory, and execute local commands (open). * Sanitization: While the skill cleans terminal artifacts and ANSI codes, it does not include sanitization or filtering to prevent the agent from following natural language instructions found in the input.

paste