The Agent Skills Directory

[SAFE]: The skill serves as a controlled orchestrator for repository maintenance tasks. It defines a rigid, sequential workflow where each stage is handled by a dedicated subagent, ensuring a structured approach to modifying the codebase.
[COMMAND_EXECUTION]: The harness explicitly manages the execution of local maintenance scripts (e.g., npm run validate, npm test, npm run evals) through a general-purpose subagent. These actions are restricted to the repository's own tooling and are intended for project validation and consistency checks.
[PROMPT_INJECTION]: The orchestrator ingests data from multiple sources including subagent summaries, repository files (skills, guides, manifests), and documentation.
Ingestion points: Reads files in skills/, evals/, and workspace notes in _workspace/brooks-harness/.
Boundary markers: Not explicitly defined in the orchestration logic; relies on task-specific subagent scoping.
Capability inventory: Spawns subagents via the Agent tool and executes shell commands for testing and validation.
Sanitization: Uses automated cross-document consistency checks and a manual feedback loop with the maintainer to validate outputs. The risk is considered low due to the closed-loop nature of the maintenance tasks on a local repository.

brooks-harness