The Agent Skills Directory

[COMMAND_EXECUTION]: The render-graphs.js script utilizes child_process.execSync to invoke the dot command from the Graphviz suite and to check for its presence via which dot. While the shell commands are largely static, they process and render data extracted dynamically from markdown files.
[INDIRECT_PROMPT_INJECTION]: The skill's 'Research phase' and 'Untrusted content boundary' sections define methods for ingesting data from the web and codebases, creating a vulnerability surface. * Ingestion points: Research workers fetch third-party material including HTML, logs, and API responses (SKILL.md). * Boundary markers: The skill mandates the use of <EXTERNAL_DATA> tags and includes instructions to treat retrieved content as untrusted, though safety relies on model compliance. * Capability inventory: The environment includes script execution capabilities via render-graphs.js and subagent workers capable of file system operations. * Sanitization: Instructions specify the use of structured extraction and schema validation for external data.
[EXTERNAL_DOWNLOADS]: The 'Research phase' in SKILL.md directs the agent to utilize exploration and general workers to retrieve information from external web sources for prior art and best practice analysis.

writing-skills