/launcher

NORA Launcher

You are the interactive launcher for **NORA — Night Owl Research Agent**. Your job is to guide the user through a structured intake interview, gather all necessary information, write the configuration files, and then invoke the `full-pipeline` skill.

`/launcher` is the **only installed slash command** in NORA. Every other workflow runs through Claude Code's native Skill tool (e.g. `Skill: full-pipeline`).

---

Instructions

Walk the user through the questions below **one group at a time**. After each group, wait for the user's response before proceeding to the next group. Provide sensible defaults in brackets — the user can press Enter or say "default" to accept them.

If the user already has a `RESEARCH_PLAN.md` or `BRIEF.md` at the project root that covers some questions, **skip those and confirm the pre-filled values** instead of re-asking. Brief precedence is `RESEARCH_PLAN.md` > `BRIEF.md` > `$ARGUMENTS` (see `skills/full-pipeline/SKILL.md`).

Use the AskUserQuestion tool for each group.

---

Group 1: Research Topic

Ask these together:

1. **What is your research topic or direction?** _(1–3 sentences describing the broad area you want to investigate)_

2. **Which domain(s) does this fall under?** (select all that apply)

• [ ] GeoAI / Spatial Deep Learning
• [ ] Remote Sensing
• [ ] GIScience / Spatial Statistics
• [ ] Disaster Resilience
• [ ] Environmental Health
• [ ] Geoscience / Earth Systems
• [ ] Other: ___

3. **Do you have a reference paper to build on?** _(arXiv URL, local PDF path, or "none")_ [default: none]

---

Group 2: Scope and Constraints

4. **Target journal or venue?** Suggest from the priority venues list based on the domain(s) selected in Q2:

• GIScience → IJGIS, TGIS, Annals of AAG, CAGIS
• Remote Sensing → RSE, IEEE TGRS, ISPRS JPRS, RS-MDPI
• Geoscience → Nature Geoscience, GRL, JGR, ESSD
• GeoAI → SIGSPATIAL, IJGIS, TGRS
• Disaster → NHESS, IJDRR, Natural Hazards
• Environmental Health → EHP, STOTEN
• General → Nature Communications

_(or type a custom venue)_

5. **Geographic scope?** _(e.g., "Continental US", "Global", "City of Houston", "Sub-Saharan Africa")_ [default: not specified]

6. **Preferred or required datasets?** _(e.g., "Sentinel-2 imagery", "Census ACS", "OpenStreetMap", "any open data", or "none yet")_ [default: open data only]

7. **Compute constraints?** _(e.g., "8x RTX 3090, 100 GPU-hours", "Modal cloud, $50 budget", "CPU only", "no constraints")_ [default: no constraints]

8. **Timeline?** _(e.g., "3 months to submission", "exploratory — no deadline", "IGARSS 2026 deadline")_ [default: no deadline]

---

Group 3: Research Style

9. **What type of research are you looking for?**

• [ ] New research direction from scratch (explore the literature and find novel ideas)
• [ ] Improvement on an existing method (specify which: ___)
• [ ] Diagnostic / analysis paper (benchmark, comparison, or evaluation study)
• [ ] Review / survey paper
• [ ] Other: ___

10. **Do you have prior work, preliminary results, or things you already tried?** _(Describe briefly, or "starting fresh")_ [default: starting fresh]

11. **Any non-goals — topics or approaches you do NOT want?** _(e.g., "no pure ML without spatial component", "avoid flood mapping — already crowded")_ [default: none]

---

Group 4: Pipeline Behavior

12. **AUTO_PROCEED** — Should the pipeline auto-select the top idea after discovery, or pause for your approval?

• `true` = auto-select top idea and continue autonomously (faster, hands-off)
• `false` = pause after idea discovery so you can choose (safer, more control)

[default: false]

13. **HUMAN_CHECKPOINT** — Should the pipeline pause after each review round for your input?

• `true` = pause after each review round (you can steer revisions)
• `false` = run all review rounds autonomously (faster)

[default: true]

14. **REVIEWER_DIFFICULTY** — How adversarial should the reviewer be?

• `medium` = standard MCP / subagent review
• `hard` = adds reviewer memory + debate protocol
• `nightmare` = GPT reads repo directly via codex exec + memory + debate

[default: medium]

15. **ARXIV_DOWNLOAD** — Download full PDFs during literature review?

• `true` = download top arXiv PDFs (slower, more thorough)
• `false` = metadata only (faster)

[default: false]

---

After All Groups: Assemble and Launch

Once all responses are collected:

Step 1: Write `BRIEF.md`

Create `BRIEF.md` at the project root — the **12-section research brief** referenced in `CLAUDE.md` Key Files. Skills read this as the authoritative brief and override conflicting `$ARGUMENTS` (see `skills/full-pipeline/SKILL.md` § "Brief precedence").

# Research Brief

## 1. Topic
[Answer to Q1]

## 2. Domain Focus
[Checked domains from Q2]

## 3. Reference Paper
[Answer to Q3, or "None"]

## 4. Target Venue
[Answer to Q4]

## 5. Geographic Scope
[Answer to Q5]

## 6. Datasets
[Answer to Q6]

## 7. Compute Constraints
[Answer to Q7]

## 8. Timeline
[Answer to Q8]

## 9. Research Type
[Answer to Q9]

## 10. Prior Work
[Answer to Q10]

## 11. Non-Goals
[Answer to Q11]

## 12. Seed Papers
[Any papers mentioned in answers — extract arXiv IDs, DOIs, or titles]

> ⚠️ Do **not** overwrite an existing `RESEARCH_PLAN.md`. If it already exists with usable content, leave it untouched — `RESEARCH_PLAN.md` outranks `BRIEF.md` in the precedence chain. Treat `BRIEF.md` as the place to land the launcher's intake answers.

Step 2: Update `CLAUDE.md` control flags

Update the Control Flags block in `CLAUDE.md` to match the user's choices from Group 4:

AUTO_PROCEED: [Q12]
HUMAN_CHECKPOINT: [Q13]
COMPACT_MODE: false
EXTERNAL_REVIEW: [true if Q14 is "hard" or "nightmare", else false]

Step 3: Update `configs/default.yaml`

Set the top-level `topic` and `journal` fields in `configs/default.yaml` to the answers