blakesims

Use when operating as the Pi architecture/design governor for the stores substrate, or when asking Pi for design review, priority coherence, contract ratification advice, or architectural approval over agent-comm.

Use when operating as the Claude Code engine controller for the stores substrate: driving tasks, daemon/worktrees, codex/rebase loops, and multi-agent coordination.

Use for the temporary front-of-engine queue-curation role in stores: triages observations/intake/tasks, deduplicates stale/noisy rows, drafts priority recommendations, and delegates backlog analysis to subagents without implementing code.

Use during wind-down when acting as engine-controller and preparing the next engine-controller agent handover.

Use for the session-scoped read-only codex sensor role in stores: watches in_review tasks, rebases review worktrees, runs codex, and posts structured review digests over agent-comm without writing substrate state or code.

Use during wind-down when acting as reviewer-runner and handing off codex/review state to the next reviewer-runner.

Use during wind-down when acting as Pi architect and handing off architectural state, priorities, and pending decisions.

Use when Blake says context is high, asks to wind down, pause/clear the engine, prepare handover, or end a multi-agent stores session cleanly. Coordinates substrate-agent/reviewer-runner feedback, updates SOP skills/docs, and produces next-session handover…

Audits CLAUDE.md files to suggest high-signal, concise rules based on recent changes or a general audit.

Independently review a debugging conclusion against risk-critical evidence; return PASS or REVISE with concrete numbered blockers, separate from suggestions.

Use whenever you are presenting the user with questions or options to reduce cognitive load and complexity. You remove the dependence to cut to the essence.

Use hypothesis-driven investigation when uncertainty could change a conclusion or safety decision. Form a testable hypothesis, run proportionate experiments, and report evidence and limitations.

Run the task-workflow development cycle — Intent Contract, routing, reviewed phased execution — whenever the user asks to use the task workflow or task system on a piece of work.

Shared task ledger, gate, artifact, and Git-safety contract for the canonical task-start workflow.

Set up the lightweight tasks/ management system in a project — directory structure, tasks/CLAUDE.md procedures, global task manager, and main.md template. Use when the user asks to set up task management or task tracking in a repository.

Shape a real task before execution. Inspect local context, suggest six deterministic inputs, call the CCA task-shape API without uploading private task context, and return the right 10-80-10 workflow, human gates, evidence requirements, recommended agent…

Guided PRD creation. Loads kit template, asks customisation questions, generates PRD and mockup at project root.

Import an existing project into the CCA workflow. Detects what you already have (PRD, code, tasks) and sets up the missing pieces so you can start building.

Welcome and onboard a new student. Set up their project with the CCA workflow, explain what's happening at every step, and guide them to their first kit.

Build your project. Orchestrates the full workflow: creates a task from your PRD, plans the implementation, then builds it phase by phase with quality checks.

Browse and download starter kits. Each kit provides a PRD template and customisation points for a specific project type.

Use whenever the user wants to update - checks for new prompts from the instructor and runs them, then offers to customize the student's setup

Run a theory-building design session against a named theory graph project using the `tg` CLI (Naur, "Programming as Theory Building"). Use when the user invokes /theory <project>, asks to record, revise or inspect the theory of a program, wants to start a new…