The Agent Skills Directory

[SAFE]: The skill establishes a framework for repository-tracking documentation (READMEs, runbooks, and agent context files) that prioritizes verifiability over narrative prose. It defines clear classes of assertions (Factual, Normative, Rationale) to ensure content remains grounded in the actual state of the repository.
[COMMAND_EXECUTION]: To ensure factual accuracy, the skill instructs agents to execute repository-local commands (e.g., npm run build, node bin/cli.js) and capture authentic output. This is a functional requirement for the skill's purpose and does not involve arbitrary or unsafe command execution.
[DATA_EXPOSURE]: The skill involves reading repository files, including source code and configuration, to generate documentation. This access is scoped to the repository being documented and is necessary for the agent to 'track the repo' as intended.
[PROMPT_INJECTION]: The instructions do not contain patterns intended to bypass safety filters, override system instructions, or extract system prompts. The guidance for AI agents ('agent-context.md') focuses on token density and relevance rather than behavioral manipulation.

writing-docs