Herd

Distill Herd commander memory, clean stale MEMORY.md entries, promote durable commander lessons into LONG_TERM_MEM.md, and propose shared-knowledge updates through a review gate. Use for periodic commander memory cleanup.

Examine workspace files for durable context, filter out low-value noise, and clean up stale memory or rule files that contribute to context rot.

Write or rewrite agent skill files so they are outcome-oriented, bounded, and easy for another agent to apply. Use when creating a new skill, translating a skill, or tightening an existing skill's acceptance criteria, boundaries, or output contract.

Add or pair a Herd machine — mint an enrollment token and return the one-line connect command, or re-pair an existing daemon machine.

Conversationally collect, validate, preview, and create a new Herd commander through the backend API.

Create a Herd automation from a conversation while preserving sourceConversationId provenance.

Turn a structured outline, report, or research brief into a presentation plan with slide-by-slide content and visual direction.

Break a large goal into ordered, non-overlapping implementation quests with dependencies and verification checks.

Prepare a concise day brief from calendar, priorities, open loops, and relevant updates.

Capture what happened, why it happened, and what should change next time.

Convert research notes and observations into reusable domain knowledge, beliefs, and follow-up questions.

Review code, architecture, plans, or pull requests for correctness, risk, missing tests, and maintainability.

Work with Google Workspace surfaces such as Gmail, Calendar, Drive, Docs, Sheets, and contacts through configured local tools.

Search local knowledge stores before duplicating research or answering from memory.

Deliver an issue end-to-end through scoped implementation, verification, and pull-request readiness.

Investigate a bug, failure, or unclear behavior and turn it into root cause plus an execution-ready issue plan.

Prepare a self-contained report artifact for publication after user approval.

Create or surface a simple timed reminder.

Run broad web or source research by splitting the question into independent lines of inquiry and synthesizing the results.