domains / research / new-project / new-project

new-project

Skill card

combined

The two skills share the name 'new-project' but perform fundamentally different tasks. The first is a software engineering workflow that initializes a project by creating a .planning/ directory structure (PROJECT.md, ROADMAP.md, etc.) and running a multi-step workflow with gates and commits. The second is a social science research workflow that conducts a structured interview to formalize a research idea, generates research questions with identification strategies, and saves a project spec to quality_reports/. They do not overlap in procedure, output, or domain, so crowning one would hide a real alternative.

Repositories

Other copies

Combined skill

# New Project

This skill covers two distinct workflows. Determine which applies based on the user's intent.

## A. Software Project Initialization

Use when the user wants to start a new software project with planning artifacts.

1. Ask configuration questions (or run in --auto mode if an idea document is provided).
2. Optionally run domain research.
3. Generate scoped requirements.
4. Build a phase-based roadmap.
5. Create the following files:
   - `.planning/PROJECT.md` — project context
   - `.planning/config.json` — workflow preferences
   - `.planning/research/` — domain research (optional)
   - `.planning/REQUIREMENTS.md` — scoped requirements
   - `.planning/ROADMAP.md` — phase structure
   - `.planning/STATE.md` — project memory
6. Preserve all workflow gates (validation, approvals, commits, routing).
7. After completion, direct the user to run the next planning phase.

## B. Research Project Formalization

Use when the user wants to develop or document a new research idea (triggers: "new project", "start research", "I have an idea", "help me formalize this idea", "what's my research question", etc.).

### Phase 1: Research Interview
- Ask one or two questions at a time; do not use AskUserQuestion — ask directly in the response.
- A good interview runs 4–6 exchanges.
- Draw out: the puzzle, why it matters, theoretical motivation, data and setting, identification strategy, expected results and contribution.
- Move to Phase 2 when you have: a clear research question, at least one plausible identification strategy, a sense of available data, and the motivation/contribution.
- If after 3 exchanges the user keeps giving vague answers, move to Phase 2 anyway and flag open questions.

### Phase 2: Research Ideation
- Generate 3–5 structured research questions ordered by type: Descriptive, Correlational, Causal, Mechanism, Policy.
- For each RQ, develop: hypothesis, identification strategy (method, treatment, control group, key assumption, robustness checks), data requirements, key pitfalls (2 threats + mitigations), and related work (2-3 papers, name only, no fabrication).
- Rank questions by feasibility × contribution.

### Phase 3: Save Project Spec
- Produce a unified project spec document.
- Save to: `quality_reports/project_spec_[sanitized_topic].md`
- Include: research question, motivation, all RQs with full detail, priority empirical strategy, and open questions.

## Conflicts

- davepoon/buildwithclaude creates `.planning/` artifacts and runs a gated workflow with commits; brycewang-stanford/Auto-Empirical-Research-Skills creates a `quality_reports/` spec and runs a conversational interview with no commit step. These are mutually exclusive output structures and must not be merged into a single file layout.