The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to use npx -y -p @javascribe/pindoupic@latest to download and execute a package from an external registry. This pattern allows for the execution of unverified remote code that can be changed at any time by the package owner. The package organization @javascribe does not match the known patterns for the specified vendor 'qxshn'.\n- [PERSISTENCE]: The install command is designed to modify host configuration files (e.g., mcp.json) and create symlinks in the ~/.agents/skills/ directory. This behavior ensures that the tool and its associated services remain active and are automatically loaded in future sessions.\n- [DATA_EXFILTRATION]: The feedback and submit_feedback commands send user-provided notes, ratings, and a 480-character 'context' snippet to the external domain pindoupic.com. This involves sending data from the local environment to a remote server not listed as a trusted vendor domain.\n- [COMMAND_EXECUTION]: The skill relies on several shell commands (npx, pindoupic, and curl) to perform its operations, providing a surface for command execution and potential argument injection.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes images directly from external URLs provided in user input. It lacks explicit sanitization or safety boundary markers for data fetched from these untrusted external sources, creating a vulnerability surface for tool output poisoning.

pindoupic-bead-pattern