The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses the scripts/fetch-guidance.sh script to download prompting guidelines from official domains belonging to OpenAI (developers.openai.com) and Anthropic (platform.claude.com). These downloads are used to provide the agent with current model optimization data. Consistent with the trust-scope-rule, these downloads from well-known service providers are documented neutrally.\n- [COMMAND_EXECUTION]: The skill executes a bundled bash script and various shell commands (such as git, ln, and fd) for repository analysis and documentation management. It includes a robust safety guard rail in SKILL.md that checks the execution environment and prevents modifications to sensitive system directories or agent configuration roots.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes repository files and external documents to generate instructions, creating a surface where content in those files could influence agent behavior.\n
Ingestion points: Processes local Markdown files (README.md, AGENTS.md, etc.) and remote prompting guides fetched from OpenAI and Anthropic.\n
Boundary markers: No explicit delimiters are used to separate ingested content from instructions.\n
Capability inventory: Authorized to write and modify Markdown files throughout the repository and execute bundled shell scripts.\n
Sanitization: The fetch-guidance.sh script performs basic validation (size and keyword presence) on downloaded content.

agents-brain