Use when the user asks to produce a PowerPoint deck — presentations, status updates, data read-outs, or structured deliverables. Takes a JSON/YAML spec and writes a .pptx with consistent slide layouts (title, section, bullets, two-column, table). Ships a runnable generate.py.
Use when the user asks to produce a formatted Excel workbook — scorecards, data tables, or multi-sheet reports — with consistent styling (headers, number formats, freeze panes, zebra striping). Ships a runnable generate.py plus a Python API.
Use when the user has an approved feature brief and wants a structured PRD written to .claude/prds/. Activates after feature-discovery completes or when the user says "write the PRD", "turn this into a PRD", "create a spec for".
Use when the user asks to build, integrate, or improve an AI/LLM feature — calling Claude, GPT, or another model API; adding prompt templates; structured output extraction; tool use; streaming; or evaluating model outputs.
Use when the user asks to add logging, metrics, tracing, health checks, or alerting to a service — or when a service lacks visibility into what it's doing in production.
Use when the user has an approved PRD and wants to decompose it into an epic with numbered task files, dependency graph, parallelization plan, and GitHub issues synced to a project board. Activates on "decompose the PRD", "create tickets", "break this into tasks", "make the epic".
Use when the user asks to implement a new feature, endpoint, service, or background job in a Python backend (FastAPI, Flask, Django, or a standalone worker). Covers the full loop: read the existing code, implement, test, and prepare for review. Trigger on "add an endpoint", "implement a service", "write a worker for…
Use when the user asks to add a component, screen, feature, or user interaction in a React + TypeScript frontend (also applicable to Vue when the project uses Vue). Covers implementation, state and data wiring, accessibility, and tests. Trigger on "add a button that…", "build the X screen", "wire this form up".
Use when the user asks to create, modify, or refactor Terraform — a new module, adding a resource to an existing module, or splitting a monolithic config into reusable modules. Produces portable, multi-cloud-aware infrastructure.
IMPORTANT: This project has a knowledge graph. ALWAYS use the code-review-graph MCP tools BEFORE using Grep/Glob/Read to explore the codebase. The graph is faster, cheaper (fewer tokens), and gives you structural context (callers, dependents, test coverage) that file scanning cannot.
Claude Code instructions for stblake/mathilda, covering workflow orchestration, 1. plan node default, 2. subagent strategy, 3. self-improvement loop and 4. verification before done.