research-experimenter

# Research Experimenter Use this skill to produce consistent, implementation-ready research artifacts. ## Required repository structure For each paper/topic: 1. Create `research/<slug>-<paper-id>/` 2. Include: - `README.md` (paper summary, extracted capability, algorithm sketch) - `reference-architecture.md` (system design, components, integration plan) - `experiments/` (implementation attempts and experiment logs) Keep `research/README.md` updated with the new paper folder. ## Stack and implementation rules - Prefer **TypeScript** for experiments and reference implementations. - Model experiments after agent-browser conventions: typed interfaces, modular components, and deterministic execution paths. - Do not default to Python unless the user explicitly asks for Python. ## Workflow 1. **Paper intake** - Capture title, canonical link, and publication date. - Write a short "what this paper proposes" section. 2. **Capability extraction** - Define one concrete capability to implement. - Explain why it matters and where it fits the product/runtime. 3. **Reference architecture** - Describe components, data flow, safety/validation gates, rollout policy, and metrics. 4. **Experiment design** - Add at least one experiment spec in `experiments/` with hypothesis, setup, procedure, acceptance criteria, and artifacts. 5. **Implementation attempt** - Add a TypeScript scaffold demonstrating the architecture in code. - Include strongly typed interfaces and minimal runnable loop logic where practical. 6. **Validation** - Run the smallest deterministic checks that prove the scaffold compiles/lints (for example, `npx tsc --noEmit` when applicable). ## Output checklist Before finalizing, confirm: - Folder created under `research/`. - `README.md`, `reference-architecture.md`, and `experiments/` exist. - At least one TypeScript implementation scaffold exists in `experiments/`. - `research/README.md` includes the paper folder.