The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various local shell commands through npm scripts, specifically npm run evals, npm run validate, and npm test as part of its QA and curation phases.
[COMMAND_EXECUTION]: The orchestrator spawns multiple sub-agents (skill-author, eval-curator, consistency-qa, trigger-boundary-auditor, release-manager) using the Agent tool to perform automated repository modifications and release management.
[COMMAND_EXECUTION]: The skill utilizes high-risk git operations including pushing directly to the main branch, potentially bypassing standard Pull Request (PR) workflows, and mentions the use of flags such as --no-verify and --force for history manipulation.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its sequential sub-agent pipeline architecture, where content or summaries generated by one agent are consumed as instructions by the next stage.
Ingestion points: Summaries stored in _workspace/brooks-harness/ and modifications made to files in skills/ and evals/ are read and processed by subsequent pipeline stages.
Boundary markers: The instructions do not define explicit boundary markers or delimiters to isolate data from instructions when passing context between sub-agents.
Capability inventory: The skill possesses capabilities for file system writes, execution of npm scripts, and authenticated git operations (push/release).
Sanitization: There is no evidence of sanitization or validation of the content passed between the author, eval, and QA stages.

brooks-harness