Composable process framework for LLM-driven work at scale: durable process record, mechanical capture & priming, honest telemetry. MCP plugin for Claude Code and compatible agents.
Agent
Surveys an existing codebase (analyze mode) or designs a new system (design mode). Returns a structured design brief as prose. Read-only — it never writes artifacts, edits code, or calls board/record tools; the invoking skill records what it returns.
Agent
Reviews a code change for correctness, security, and quality. Returns severity-classified findings. Read-only — no MCP tools; reads board/record evidence via the ideate CLIs; the invoking skill records the findings it accepts.
Agent
Breaks a goal or design into a set of independently-executable work items with explicit dependency and containment edges. Returns a structured item list as JSON; the invoking skill creates the board items. Read-only.
Agent
Keeps the steering store (guiding principles, constraints, policies) coherent — spotting contradictions, redundancy, and drift, and proposing precise amendments. Returns proposed steering changes; the invoking skill applies them via steeringput.
Agent
Finds what is missing rather than what is wrong — unhandled cases, absent tests, unaddressed requirements, second consumers of a changed interface. Returns gap findings. Read-only; the invoking skill records them.
Agent
Composes a concise, recall-shaped narrative of what happened in a cycle or session — decisions, findings, what was built, what's next. Returns the prose; the invoking skill appends it to the process record.
Agent
Stands in for the user during unattended autopilot runs — makes the judgment call an Andon escalation would otherwise interrupt a human for, within the project's stated principles and appetite. Returns a decision with rationale; autopilot records it.
Agent
Investigates an open question by reading code, docs, and the web, and returns concise sourced findings with a recommendation. Read-only — returns notes; the invoking skill decides what to record.
Agent
Checks whether completed work actually satisfies its spec, the guiding principles, and the policies in steering. Returns adherence findings. Read-only; the invoking skill records them.
Agent
Implements exactly one board work item to completion — writes code, runs tests, verifies. The only agent with edit tools. Returns a completion report; the invoking skill owns the claim/complete board transitions.