Night Owl Research Agent

GRIND-Lab-Core107stars

Fully automatic AI research agent for Geoscientists, Remote Sensing researchers, and GIScientists. Features harness engineering, GeoBenchmark (OLS/GWR/MGWR), journal templates, and MCP servers.

Install Run on PlugMCP

GitHub

Run on PlugMCP

1# Night Owl Research Agent

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
Last checked: 2026-10-28

Stars: 107
Tools: —
Uptime: —

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Night Owl Research Agent?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Night Owl Research Agent need an API key?

No required environment variable is listed on this record.

What transport does Night Owl Research Agent use?

Night Owl Research Agent runs locally over stdio.

When was Night Owl Research Agent last checked?

Last checked 2026-10-28. Fully automatic AI research agent for Geoscientists, Remote Sensing researchers, and GIScientists. Features harness engineering, GeoBenchmark (OLS/GWR/MGWR), journal templates, and MCP servers.

From the publisher

Documentation

NORA (Night Owl Research Agent)

A fully automatic, domain-aware AI research agent for Geoscientists, Remote Sensing researchers, and GIScientists — powered entirely by Claude Code skills.

NORA Banner

Quick Start

NORA runs inside Claude Code. There is no Python entry point, no server to spin up, and no build step — you just drop the skills into Claude Code's skill directory and invoke the launcher.

Step 1 — Install Claude Code

Install Claude Code first. Any of the official distributions works:

CLI (recommended):

npm install -g @anthropic-ai/claude-code
claude --version

Desktop app (macOS / Windows): download from https://claude.com/claude-code
VS Code extension: install "Claude Code" from the Marketplace
Web: https://claude.ai/code

Step 2 — Get NORA onto your machine

git clone https://github.com/GRIND-Lab-Core/night_owl_research_agent.git
cd night_owl_research_agent

Step 3 — Install the skills into Claude Code

Claude Code looks for skills under ~/.claude/skills/ (user-level, available in every project) or <project>/.claude/skills/ (project-local). Copy the entire skills/ folder from this repo into one of those locations.

macOS / Linux (user-level — recommended):

mkdir -p ~/.claude/skills
cp -R skills/* ~/.claude/skills/

Windows PowerShell (user-level):

New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.claude\skills" | Out-Null
Copy-Item -Recurse -Force .\skills\* "$env:USERPROFILE\.claude\skills\"

Windows bash / Git Bash:

mkdir -p "$USERPROFILE/.claude/skills"
cp -R skills/* "$USERPROFILE/.claude/skills/"

Project-local alternative (skills only visible when Claude Code is opened in this folder):

mkdir -p .claude/skills
cp -R skills/* .claude/skills/

Also copy the launcher slash command so /launcher is available:

# macOS / Linux
mkdir -p ~/.claude/commands
cp .claude/commands/launcher.md ~/.claude/commands/

# Windows PowerShell
New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.claude\commands" | Out-Null
Copy-Item -Force .\.claude\commands\launcher.md "$env:USERPROFILE\.claude\commands\"

Verify the install — open Claude Code and run:

/skills

You should see the NORA skills (full-pipeline, lit-review, idea-discovery-pipeline, deploy-experiment, paper-draft, …) listed.

Step 4 — Start a research session

Open the night_owl_research_agent folder in Claude Code (this gives NORA access to CLAUDE.md, RESEARCH_PLAN.md, output/, memory/, and tools/), then pick one of the two entry points:

Option A — interactive launcher (best for first-time users):

/launcher

The launcher walks you through a short questionnaire — research topic, stage to start from, control flags (AUTO_PROCEED, HUMAN_CHECKPOINT, COMPACT_MODE, REVIEWER_DIFFICULTY) — and routes to the correct skill.

Option B — end-to-end pipeline (best when you already know what you want to run):

Skill: full-pipeline
"Your research direction here, e.g. 'urban soundscape inequality via street-view + audio foundation models'"

or, if you prefer a slash-style invocation:

/full-pipeline "your research direction"

full-pipeline chains all four stages: idea-discovery-pipeline → deploy-experiment → auto-review-loop → generate-report and then hands off to paper-writing-pipeline for the manuscript.

Tip: for reproducibility, fill in RESEARCH_PLAN.md (or BRIEF.md) in the project root before launching. When either file is present, skills read it as the authoritative brief and ignore conflicting $ARGUMENTS.

Step 5 — (Optional) Enable extras

MCP servers — edit .mcp.json and register with Claude Code (/mcp inside the chat) to enable filesystem, fetch, arxiv_mcp, geo_mcp, github, and brave_search. See mcp/README_MCP.md.
Hooks — settings.json wires harness/hooks/*.sh into Claude Code's lifecycle (writes handoff.json on session end, validates tool use, sends desktop notifications). On Windows, run the hook scripts via Git Bash or WSL.
W&B — if your experiments use Weights & Biases, run wandb login once on the host where deploy-experiment will launch training.
API keys — set ANTHROPIC_API_KEY (for Claude Code), plus any optional keys you want to use (SEMANTIC_SCHOLAR_API_KEY, GITHUB_TOKEN, BRAVE_API_KEY).

Prerequisites summary

Requirement	Why
Claude Code (CLI / desktop / web / VS Code)	Runtime for skills
Anthropic account + API credit	Powers the agent
Python 3.10+ with `pip install arxiv requests`	`tools/arxiv_fetch.py`, `tools/semantic_scholar_fetch.py`
Conda env with `geopandas, pysal, libpysal, esda, spreg, mgwr, rasterio, xarray`	Track B (spatial) experiments
CUDA GPU (local / remote SSH / Modal)	Track A (deep-learning) experiments — optional

What It Does

NORA automates the complete academic research lifecycle using Claude Code skills — Markdown-defined workflows that Claude reads and executes, selecting appropriate tools and methods based on context.

Literature review — searches ArXiv, Semantic Scholar, local papers, Zotero, and Obsidian; synthesizes findings and identifies ranked research gaps.
Idea discovery — generates 8–12 research ideas from literature gaps, validates novelty via multi-source search + external reviewer, and pilot-tests the top candidates. Pilots run a mandatory local-GPU presence check first (nvidia-smi → CUDA, then MPS, then none); when a local GPU is detected, every pilot launches on it instead of silently falling back to CPU or remote.
Method refinement — iteratively refines vague research directions into problem-anchored, implementation-ready proposals via adversarial review (up to 5 rounds, score ≥ 9 target).
Experiment design & execution — produces claim-driven experiment roadmaps and deploys to local, remote SSH, or Modal serverless GPU (Track A), or runs spatial/GIScience methods on CPU (Track B), or both for mixed GeoAI. The same mandatory local-GPU check runs at Step 0 of deploy-experiment so any ML/DL workload (pilot or full) executes on the local GPU when present.
Data acquisition — discovers, evaluates, downloads, validates, and documents datasets from government portals, APIs, cloud archives, and open repositories with full provenance.
Spatial analysis — guideline-driven: classifies the analytical objective, runs ESDA, and applies geospatial diagnostics conditionally on the research question. MAUP discussion, GWR/MGWR, residual Moran's I, alternative spatial weights, and spatial CV are triggered only when the claim depends on them; when in doubt the skill pauses for a human checkpoint instead of running heavyweight checks (or skipping reviewer-expected ones) by reflex.
Adversarial review — up to 4 rounds of generator–evaluator-separated review with per-criterion hard floors; medium / hard / nightmare reviewer modes via Codex MCP, codex exec, or a Claude subagent. Domain personas (giscience, remote-sensing, spatial-data-science) apply geo-specific must-checks only where the paper's claims actually depend on them instead of penalizing every paper for missing MAUP / GWR discussion.
Report + paper writing — consolidates every pipeline artifact into output/NARRATIVE_REPORT.md, then runs paper-writing-pipeline to produce a journal-ready manuscript (Markdown → LaTeX → PDF/DOCX) with journal-specific profiles for IJGIS, IEEE TGRS, ISPRS JPRS, RSE, AAG, TGIS, and more.

Architecture

NORA is a skills-first system. All research logic lives in Markdown skill files that Claude reads and executes.

NORA Architecture

Skills describe workflow logic in Markdown. Claude reads a skill to understand the workflow, then decides the exact sequence of actions based on context — the skill provides guidelines and decision frameworks, not rigid procedures.

You (or /launcher)
    ↓ invokes
Skill SKILL.md  ←─── reads domain knowledge from skills/knowledge/
    ↓ Claude decides what to do
CLI tools (tools/arxiv_fetch.py, etc.) + inline Python + MCP servers as needed
    ↓ produce
Output files (reports, paper-cache, figures, manuscript)
    ↓ read by
Next skill in pipeline

The single installed slash command is /launcher. Every other skill is invoked by name (Claude Code's native Skill tool) or by being called internally from another skill.

Skills

23 workflow skills in skills/ plus domain knowledge in skills/knowledge/. Each skill is a self-contained Markdown workflow file.

Workflow Skills

Skill	What it does
`full-pipeline`	Master pipeline: idea discovery → experiment → review → report → paper
`lit-review`	Search + synthesize + gap analysis (ArXiv, Semantic Scholar, local papers, Zotero, Obsidian)
`idea-discovery-pipeline`	Full idea pipeline: lit-review → generate-idea → novelty-check → idea-review → experiment-design-pipeline
`generate-idea`	Brainstorm 8–12 ideas, filter, pilot-test top 3, rank (called by idea-discovery-pipeline)
`novelty-check`	Verify idea novelty via multi-source search + external reviewer
`idea-review`	External critical review of research ideas (Codex MCP)
`refine-research`	Iterative method refinement via external review (up to 5 rounds, score ≥ 9)
`experiment-design`	Claim-driven experiment roadmap with run order, budget, decision gates
`experiment-design-pipeline`	One-shot wrapper: refine-research → experiment-design
`deploy-experiment`	Deploy experiments — mandatory local-GPU check at Step 0, then Track A (GPU ML) and/or Track B (CPU spatial)
`data-download`	Discover, evaluate, download datasets with provenance tracking
`spatial-analysis`	Research-question-driven spatial analysis: classification → ESDA → method → conditional diagnostics → interpretation, with a human checkpoint before adding or skipping heavyweight spatial checks
`auto-review-loop`	Up to 4 adversarial review rounds with per-criterion floors
`generate-report`	Consolidate lit-review + idea + experiment + review artifacts into `output/NARRATIVE_REPORT.md`
`paper-writing-pipeline`	Orchestrates paper-plan → paper-figure-generate → paper-draft → paper-review-loop → paper-covert
`paper-plan`	Build section outline + figure plan (`output/PAPER_PLAN.md`)
`paper-figure-generate`	Generate publication-quality figures, maps, diagrams, and captions
`paper-draft`	Turn `output/PAPER_PLAN.md` into a journal-quality Markdown manuscript
`paper-review-loop`	Reviewer-editor review of the draft manuscript and iterative revision
`paper-covert`	Convert final manuscript into venue submission package (modular LaTeX, PDF, DOCX)
`submit-check`	Validate manuscript against target-journal requirements
`training-check`	Monitor running experiments for stalls/failures

Domain Knowledge

File	Domain
`spatial-methods.md`	Spatial statistics, regression, autocorrelation
`geoai-domain.md`	GeoAI, spatial deep learning, foundation models
`academic-writing.md`	Academic writing conventions
`apa-citations.md`	APA 7th edition citation formatting
`disaster-resilience.md`	Disaster management, community resilience
`environmental-health.md`	Environmental epidemiology, exposure assessment
`literature-mining.md`	Literature search and synthesis strategies
`research-iteration.md`	Iterative research refinement patterns

Control Flags

Edit CLAUDE.md before starting a long run:

AUTO_PROCEED: false       # true = auto-select top idea after discovery; false = wait for approval
HUMAN_CHECKPOINT: true    # true = pause after each review round; false = run autonomously
COMPACT_MODE: false       # true = use output/PROJ_NOTES.md instead of full logs (saves context)
EXTERNAL_REVIEW: false    # true = use Claude subagent / external reviewer LLM

full-pipeline also accepts REVIEWER_DIFFICULTY = medium | hard | nightmare and ARXIV_DOWNLOAD = true | false. Overrides can be passed inline, e.g.:

/full-pipeline "topic — AUTO_PROCEED: false, difficulty: nightmare"

Harness Engineering

Claude Code's hook system automates lifecycle management (configured in settings.json):

Hook	When	What it does
`PreToolUse`	Before Bash/Write	Validates paths, blocks dangerous commands, logs intent
`PostToolUse`	After tool execution	Updates state, caches results
`SkillUse`	Before/after each Skill tool call	`harness/hooks/skill_marker.sh` writes per-stage markers feeding `tools/telemetry_stage_marker.py`
`Stop`	Agent session ends	Writes `handoff.json`, updates `memory/MEMORY.md`, runs `tools/telemetry_aggregate.py` to emit `output/TELEMETRY.jsonl` and `output/TELEMETRY_STAGES.jsonl`, sends notification
`Notification`	Long tasks finish	Desktop alert via `notify-send` / `osascript`

Autoresearch Scoring Loop

paper-draft writes draft
    ↓
paper-review-loop scores it (separate context — generator–evaluator separation)
    ↓
All 5 dimension floors met AND weighted avg ≥ 7.5? → ACCEPT
    ↓ (else)
paper-draft revises (max 3 attempts total)
    ↓
If still not accepted → flag for human review

Dimension	Weight	Hard floor
Novelty	30%	≥ 6.5
Rigor	25%	≥ 7.0
Literature coverage	20%	≥ 6.5
Clarity	15%	≥ 6.0
Impact	10%	≥ 6.0

Accept requires weighted avg ≥ 7.5 and all five floors met.

Journal Templates & Profiles

Templates enforce correct structure, section ordering, word limits, and formatting. paper-covert additionally loads a YAML profile that drives LaTeX conversion.

Markdown templates (`templates/`)

Category	Journals
`geoscience/`	Nature Geoscience, Geophysical Research Letters
`remote_sensing/`	Remote Sensing of Environment, IEEE TGRS, ISPRS JPRS
`giscience/`	IJGIS, Transactions in GIS, Annals of AAG

Submission profiles (`skills/paper-covert/profiles/`)

aag_annals.yaml, generic.yaml, ieee_tgrs.yaml, ijgis.yaml, isprs_jprs.yaml, rse.yaml, tgis.yaml.

MCP Servers

Declared in .mcp.json. Setup notes in mcp/README_MCP.md.

Server	Purpose
`filesystem`	Read/write local files and datasets
`fetch`	Fetch web content (papers, data portals, journal pages)
`geo_mcp`	Spatial data: GADM, OSM Overpass, Census ACS, GEE (`mcp/geo_mcp_server.py`)
`arxiv_mcp`	ArXiv search, paper fetch, abstract parsing
`github`	GitHub repo reading and code management
`brave_search`	Web search for literature, datasets, documentation

Key Output Files

File	Written by
`output/LIT_REVIEW_REPORT.md`	`lit-review`
`output/IDEA_REPORT.md` / `NOVELTY_REPORT.md` / `IDEA_REVIEW_REPORT.md`	`idea-discovery-pipeline`
`output/refine-logs/FINAL_PROPOSAL.md` / `REFINE_REPORT.md`	`refine-research`
`output/refine-logs/EXPERIMENT_PLAN.md` / `output/EXPERIMENT_TRACKER.md`	`experiment-design`
`output/experiment/EXPERIMENT_RESULT.md` / `EXPERIMENT_LOG.md`	`deploy-experiment`
`output/experiment/data/` / `figures/` / `scripts/`	`deploy-experiment`, `spatial-analysis`
`output/AUTO_REVIEW_REPORT.md` / `REVIEW_STATE.json` / `review-rounds/`	`auto-review-loop`
`output/METHOD_DESCRIPTION.md`	`auto-review-loop`
`output/NARRATIVE_REPORT.md`	`generate-report`
`output/PAPER_PLAN.md`	`paper-plan`
`output/figures/`	`paper-figure-generate`
`output/manuscript/`	`paper-draft`, `paper-review-loop`
`output/papers/`	`paper-covert`
`output/reports/submit_check_*.md`	`submit-check`
`data/DATA_MANIFEST.md`, `data/raw/`	`data-download`
`output/PROJ_NOTES.md`	all skills (append-only, compact log)
`output/TELEMETRY.jsonl` (per-session) and `output/TELEMETRY_STAGES.jsonl` (per-skill)	`tools/telemetry_aggregate.py` (run by Stop hook)
`output/CONTRACT_VIOLATION.md`	any skill that detects a downgraded success criterion or other contract violation
`memory/MEMORY.md`, `handoff.json`	Stop hook

Project Structure

night_owl_research_agent/
├── CLAUDE.md                        ← Dashboard and project conventions
├── README.md                        ← This file
├── design_principle.md              ← Skill-level design principles (export → Excel via tools/)
├── design_principle_agents.md       ← Sub-agent design principles
├── settings.json                    ← Claude Code hooks, permissions, env vars
├── .mcp.json                        ← MCP server declarations
│
├── .claude/
│   ├── commands/
│   │   └── launcher.md              ← /launcher (only installed slash command)
│   └── agents/                      ← Specialist sub-agent definitions (9 total)
│       ├── orchestrator.md
│       ├── literature-scout.md
│       ├── synthesis-analyst.md
│       ├── gap-finder.md
│       ├── hypothesis-generator.md
│       ├── geo-specialist.md
│       ├── paper-writer.md
│       ├── peer-reviewer.md
│       └── citation-manager.md
│
├── skills/                          ← 22 workflow skills + knowledge/
│   ├── full-pipeline/SKILL.md
│   ├── lit-review/SKILL.md
│   ├── idea-discovery-pipeline/SKILL.md
│   ├── generate-idea/SKILL.md
│   ├── novelty-check/SKILL.md
│   ├── idea-review/SKILL.md
│   ├── refine-research/SKILL.md
│   ├── experiment-design/SKILL.md
│   ├── experiment-design-pipeline/SKILL.md
│   ├── deploy-experiment/SKILL.md
│   ├── data-download/SKILL.md
│   ├── spatial-analysis/SKILL.md
│   ├── auto-review-loop/SKILL.md
│   ├── generate-report/{SKILL.md, templates/}
│   ├── paper-writing-pipeline/SKILL.md
│   ├── paper-plan/SKILL.md
│   ├── paper-figure-generate/{SKILL.md, templates/}
│   ├── paper-draft/{SKILL.md, templates/}
│   ├── paper-review-loop/{SKILL.md, templates/}
│   ├── paper-covert/{SKILL.md, profiles/, templates/}
│   ├── submit-check/SKILL.md
│   ├── training-check/SKILL.md
│   └── knowledge/                   ← Domain reference files
│
├── tools/                           ← CLI utilities (called by skills + harness)
│   ├── arxiv_fetch.py
│   ├── semantic_scholar_fetch.py
│   ├── convert_skills_to_llm_chat.py
│   ├── export_design_principle_table.py        ← exports design_principle.md tables to Excel
│   ├── export_agent_design_principle_table.py  ← exports design_principle_agents.md tables
│   ├── telemetry_stage_marker.py               ← called by skill_marker hook (per-skill timing)
│   └── telemetry_aggregate.py                  ← called by Stop hook (session/stage telemetry)
│
├── configs/
│   └── default.yaml                 ← Scoring weights, domain keywords
│
├── templates/                       ← Project + paper templates
│   ├── EXPERIMENT_LOG_TEMPLATE.md
│   ├── EXPERIMENT_PLAN_TEMPLATE.md
│   ├── FINDINGS_TEMPLATE.md
│   ├── HANDOFF_TEMPLATE.json
│   ├── IDEA_CANDIDATES_TEMPLATE.md
│   ├── PAPER_PLAN_TEMPLATE.md
│   ├── RESEARCH_CONTRACT_TEMPLATE.md
│   ├── RESEARCH_PLAN_TEMPLATE.md
│   ├── REVIEW_STATE_TEMPLATE.json
│   ├── geoscience/ (nature_geoscience, grl_template)
│   ├── remote_sensing/ (ieee_tgrs, isprs_jprs, remote_sensing_env)
│   └── giscience/ (ijgis, transactions_gis, annals_aag)
│
├── harness/
│   ├── hooks/ (pre_tool_use, post_tool_use, skill_marker, stop_hook, notification)
│   └── prompts/system_geo.md
│
├── mcp/                             ← MCP server implementations
│   ├── geo_mcp_server.py
│   └── README_MCP.md
│
├── memory/MEMORY.md                 ← Persistent session memory
│
├── output/                          ← All generated outputs
│   ├── AUTO_REVIEW.md
│   ├── REVIEW_STATE.json
│   ├── ARCHITECTURE_DIAGRAM_PROMPTS.md
│   ├── papers/
│   ├── figures/
│   └── reports/
│
├── res/nora_architecture.png        ← Architecture diagram
│
└── archived/                        ← Retired skills and pre-skill Python modules

Contributing

Fork this repository.
Add skills in skills/<name>/SKILL.md.
Add journal templates in templates/ (plus a YAML profile in skills/paper-covert/profiles/ if needed).
Add domain knowledge in skills/knowledge/.

License

MIT License. See LICENSE for details.

Design Principles

Two living documents describe the rules NORA's skills and sub-agents follow. Treat them as the source of truth when you write or change a skill.

File	Scope
`design_principle.md`	Skill-level principles: anchored problem, smallest adequate mechanism, generator–evaluator separation, conditional geospatial checks, mandatory local-GPU check before any pilot or full experiment, human-checkpoint pattern for synthesis decisions.
`design_principle_agents.md`	Sub-agent principles for the 9 specialists in `.claude/agents/` (orchestrator, literature-scout, gap-finder, hypothesis-generator, geo-specialist, paper-writer, peer-reviewer, citation-manager, synthesis-analyst).

Both files can be exported to Excel for review or workshop use:

python tools/export_design_principle_table.py
python tools/export_agent_design_principle_table.py

Inspired By

NORA's design borrows ideas from several open-source projects. Credit and gratitude to their authors:

BZBarrett/superpowers — skill-pack patterns for extending Claude Code with composable Markdown workflows.
BZBarrett/get-shit-done — pragmatic harness patterns for getting long-running agentic work to actually finish.
wanshuiyin/Auto-claude-code-research-in-sleep — "research while you sleep" autonomous-loop concept that motivated NORA's overnight pipelines, handoff.json recovery, and adversarial review loop.
karpathy/autoresearch — generator–evaluator separation and the per-criterion floors + weighted-average scoring loop adapted into auto-review-loop and paper-review-loop.

If your project influenced NORA and is missing here, please open an issue and we will add it.

Citation

If you use NORA in your research, please cite the arXiv preprint:

Zhou, B., Wu, Q., Huang, X., Ning, H., Li, D., & Zhang, Z. (2026). NORA: Night Owl Research Agent — Autonomous AI Research for Geoscience, Remote Sensing, and GIScience. arXiv:2605.02092. https://arxiv.org/abs/2605.02092

@misc{zhou2026nora,
  title         = {NORA: Night Owl Research Agent --- Autonomous AI Research for Geoscience, Remote Sensing, and GIScience},
  author        = {Zhou, Bing and Wu, Qiusheng and Huang, Xiao and Ning, Huan and Li, Diya and Zhang, Ziyi},
  year          = {2026},
  eprint        = {2605.02092},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2605.02092},
  howpublished  = {\url{https://github.com/GRIND-Lab-Core/night_owl_research_agent}}
}

Night Owl Research Agent

GRIND-Lab-Core107stars

Fully automatic AI research agent for Geoscientists, Remote Sensing researchers, and GIScientists. Features harness engineering, GeoBenchmark (OLS/GWR/MGWR), journal templates, and MCP servers.

Install Run on PlugMCP

GitHub

Run on PlugMCP

1# Night Owl Research Agent

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
Last checked: 2026-10-28

Stars: 107
Tools: —
Uptime: —

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Night Owl Research Agent?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Night Owl Research Agent need an API key?

No required environment variable is listed on this record.

What transport does Night Owl Research Agent use?

Night Owl Research Agent runs locally over stdio.

When was Night Owl Research Agent last checked?

From the publisher

Documentation

NORA (Night Owl Research Agent)

A fully automatic, domain-aware AI research agent for Geoscientists, Remote Sensing researchers, and GIScientists — powered entirely by Claude Code skills.

NORA Banner

Quick Start

NORA runs inside Claude Code. There is no Python entry point, no server to spin up, and no build step — you just drop the skills into Claude Code's skill directory and invoke the launcher.

Step 1 — Install Claude Code

Install Claude Code first. Any of the official distributions works:

CLI (recommended):

npm install -g @anthropic-ai/claude-code
claude --version

Desktop app (macOS / Windows): download from https://claude.com/claude-code
VS Code extension: install "Claude Code" from the Marketplace
Web: https://claude.ai/code

Step 2 — Get NORA onto your machine

git clone https://github.com/GRIND-Lab-Core/night_owl_research_agent.git
cd night_owl_research_agent

Step 3 — Install the skills into Claude Code

macOS / Linux (user-level — recommended):

mkdir -p ~/.claude/skills
cp -R skills/* ~/.claude/skills/

Windows PowerShell (user-level):

New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.claude\skills" | Out-Null
Copy-Item -Recurse -Force .\skills\* "$env:USERPROFILE\.claude\skills\"

Windows bash / Git Bash:

mkdir -p "$USERPROFILE/.claude/skills"
cp -R skills/* "$USERPROFILE/.claude/skills/"

Project-local alternative (skills only visible when Claude Code is opened in this folder):

mkdir -p .claude/skills
cp -R skills/* .claude/skills/

Also copy the launcher slash command so /launcher is available:

# macOS / Linux
mkdir -p ~/.claude/commands
cp .claude/commands/launcher.md ~/.claude/commands/

# Windows PowerShell
New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.claude\commands" | Out-Null
Copy-Item -Force .\.claude\commands\launcher.md "$env:USERPROFILE\.claude\commands\"

Verify the install — open Claude Code and run:

/skills

You should see the NORA skills (full-pipeline, lit-review, idea-discovery-pipeline, deploy-experiment, paper-draft, …) listed.

Step 4 — Start a research session

Open the night_owl_research_agent folder in Claude Code (this gives NORA access to CLAUDE.md, RESEARCH_PLAN.md, output/, memory/, and tools/), then pick one of the two entry points:

Option A — interactive launcher (best for first-time users):

/launcher

Option B — end-to-end pipeline (best when you already know what you want to run):

Skill: full-pipeline
"Your research direction here, e.g. 'urban soundscape inequality via street-view + audio foundation models'"

or, if you prefer a slash-style invocation:

/full-pipeline "your research direction"

full-pipeline chains all four stages: idea-discovery-pipeline → deploy-experiment → auto-review-loop → generate-report and then hands off to paper-writing-pipeline for the manuscript.

Step 5 — (Optional) Enable extras

MCP servers — edit .mcp.json and register with Claude Code (/mcp inside the chat) to enable filesystem, fetch, arxiv_mcp, geo_mcp, github, and brave_search. See mcp/README_MCP.md.
Hooks — settings.json wires harness/hooks/*.sh into Claude Code's lifecycle (writes handoff.json on session end, validates tool use, sends desktop notifications). On Windows, run the hook scripts via Git Bash or WSL.
W&B — if your experiments use Weights & Biases, run wandb login once on the host where deploy-experiment will launch training.
API keys — set ANTHROPIC_API_KEY (for Claude Code), plus any optional keys you want to use (SEMANTIC_SCHOLAR_API_KEY, GITHUB_TOKEN, BRAVE_API_KEY).

Prerequisites summary

Requirement	Why
Claude Code (CLI / desktop / web / VS Code)	Runtime for skills
Anthropic account + API credit	Powers the agent
Python 3.10+ with `pip install arxiv requests`	`tools/arxiv_fetch.py`, `tools/semantic_scholar_fetch.py`
Conda env with `geopandas, pysal, libpysal, esda, spreg, mgwr, rasterio, xarray`	Track B (spatial) experiments
CUDA GPU (local / remote SSH / Modal)	Track A (deep-learning) experiments — optional

What It Does

Literature review — searches ArXiv, Semantic Scholar, local papers, Zotero, and Obsidian; synthesizes findings and identifies ranked research gaps.
Idea discovery — generates 8–12 research ideas from literature gaps, validates novelty via multi-source search + external reviewer, and pilot-tests the top candidates. Pilots run a mandatory local-GPU presence check first (nvidia-smi → CUDA, then MPS, then none); when a local GPU is detected, every pilot launches on it instead of silently falling back to CPU or remote.
Method refinement — iteratively refines vague research directions into problem-anchored, implementation-ready proposals via adversarial review (up to 5 rounds, score ≥ 9 target).
Experiment design & execution — produces claim-driven experiment roadmaps and deploys to local, remote SSH, or Modal serverless GPU (Track A), or runs spatial/GIScience methods on CPU (Track B), or both for mixed GeoAI. The same mandatory local-GPU check runs at Step 0 of deploy-experiment so any ML/DL workload (pilot or full) executes on the local GPU when present.
Data acquisition — discovers, evaluates, downloads, validates, and documents datasets from government portals, APIs, cloud archives, and open repositories with full provenance.
Spatial analysis — guideline-driven: classifies the analytical objective, runs ESDA, and applies geospatial diagnostics conditionally on the research question. MAUP discussion, GWR/MGWR, residual Moran's I, alternative spatial weights, and spatial CV are triggered only when the claim depends on them; when in doubt the skill pauses for a human checkpoint instead of running heavyweight checks (or skipping reviewer-expected ones) by reflex.
Adversarial review — up to 4 rounds of generator–evaluator-separated review with per-criterion hard floors; medium / hard / nightmare reviewer modes via Codex MCP, codex exec, or a Claude subagent. Domain personas (giscience, remote-sensing, spatial-data-science) apply geo-specific must-checks only where the paper's claims actually depend on them instead of penalizing every paper for missing MAUP / GWR discussion.
Report + paper writing — consolidates every pipeline artifact into output/NARRATIVE_REPORT.md, then runs paper-writing-pipeline to produce a journal-ready manuscript (Markdown → LaTeX → PDF/DOCX) with journal-specific profiles for IJGIS, IEEE TGRS, ISPRS JPRS, RSE, AAG, TGIS, and more.

Architecture

NORA is a skills-first system. All research logic lives in Markdown skill files that Claude reads and executes.

NORA Architecture

You (or /launcher)
    ↓ invokes
Skill SKILL.md  ←─── reads domain knowledge from skills/knowledge/
    ↓ Claude decides what to do
CLI tools (tools/arxiv_fetch.py, etc.) + inline Python + MCP servers as needed
    ↓ produce
Output files (reports, paper-cache, figures, manuscript)
    ↓ read by
Next skill in pipeline

The single installed slash command is /launcher. Every other skill is invoked by name (Claude Code's native Skill tool) or by being called internally from another skill.

Skills

23 workflow skills in skills/ plus domain knowledge in skills/knowledge/. Each skill is a self-contained Markdown workflow file.

Workflow Skills

Skill	What it does
`full-pipeline`	Master pipeline: idea discovery → experiment → review → report → paper
`lit-review`	Search + synthesize + gap analysis (ArXiv, Semantic Scholar, local papers, Zotero, Obsidian)
`idea-discovery-pipeline`	Full idea pipeline: lit-review → generate-idea → novelty-check → idea-review → experiment-design-pipeline
`generate-idea`	Brainstorm 8–12 ideas, filter, pilot-test top 3, rank (called by idea-discovery-pipeline)
`novelty-check`	Verify idea novelty via multi-source search + external reviewer
`idea-review`	External critical review of research ideas (Codex MCP)
`refine-research`	Iterative method refinement via external review (up to 5 rounds, score ≥ 9)
`experiment-design`	Claim-driven experiment roadmap with run order, budget, decision gates
`experiment-design-pipeline`	One-shot wrapper: refine-research → experiment-design
`deploy-experiment`	Deploy experiments — mandatory local-GPU check at Step 0, then Track A (GPU ML) and/or Track B (CPU spatial)
`data-download`	Discover, evaluate, download datasets with provenance tracking
`spatial-analysis`	Research-question-driven spatial analysis: classification → ESDA → method → conditional diagnostics → interpretation, with a human checkpoint before adding or skipping heavyweight spatial checks
`auto-review-loop`	Up to 4 adversarial review rounds with per-criterion floors
`generate-report`	Consolidate lit-review + idea + experiment + review artifacts into `output/NARRATIVE_REPORT.md`
`paper-writing-pipeline`	Orchestrates paper-plan → paper-figure-generate → paper-draft → paper-review-loop → paper-covert
`paper-plan`	Build section outline + figure plan (`output/PAPER_PLAN.md`)
`paper-figure-generate`	Generate publication-quality figures, maps, diagrams, and captions
`paper-draft`	Turn `output/PAPER_PLAN.md` into a journal-quality Markdown manuscript
`paper-review-loop`	Reviewer-editor review of the draft manuscript and iterative revision
`paper-covert`	Convert final manuscript into venue submission package (modular LaTeX, PDF, DOCX)
`submit-check`	Validate manuscript against target-journal requirements
`training-check`	Monitor running experiments for stalls/failures

Domain Knowledge

File	Domain
`spatial-methods.md`	Spatial statistics, regression, autocorrelation
`geoai-domain.md`	GeoAI, spatial deep learning, foundation models
`academic-writing.md`	Academic writing conventions
`apa-citations.md`	APA 7th edition citation formatting
`disaster-resilience.md`	Disaster management, community resilience
`environmental-health.md`	Environmental epidemiology, exposure assessment
`literature-mining.md`	Literature search and synthesis strategies
`research-iteration.md`	Iterative research refinement patterns

Control Flags

Edit CLAUDE.md before starting a long run:

AUTO_PROCEED: false       # true = auto-select top idea after discovery; false = wait for approval
HUMAN_CHECKPOINT: true    # true = pause after each review round; false = run autonomously
COMPACT_MODE: false       # true = use output/PROJ_NOTES.md instead of full logs (saves context)
EXTERNAL_REVIEW: false    # true = use Claude subagent / external reviewer LLM

full-pipeline also accepts REVIEWER_DIFFICULTY = medium | hard | nightmare and ARXIV_DOWNLOAD = true | false. Overrides can be passed inline, e.g.:

/full-pipeline "topic — AUTO_PROCEED: false, difficulty: nightmare"

Harness Engineering

Claude Code's hook system automates lifecycle management (configured in settings.json):

Hook	When	What it does
`PreToolUse`	Before Bash/Write	Validates paths, blocks dangerous commands, logs intent
`PostToolUse`	After tool execution	Updates state, caches results
`SkillUse`	Before/after each Skill tool call	`harness/hooks/skill_marker.sh` writes per-stage markers feeding `tools/telemetry_stage_marker.py`
`Stop`	Agent session ends	Writes `handoff.json`, updates `memory/MEMORY.md`, runs `tools/telemetry_aggregate.py` to emit `output/TELEMETRY.jsonl` and `output/TELEMETRY_STAGES.jsonl`, sends notification
`Notification`	Long tasks finish	Desktop alert via `notify-send` / `osascript`

Autoresearch Scoring Loop

paper-draft writes draft
    ↓
paper-review-loop scores it (separate context — generator–evaluator separation)
    ↓
All 5 dimension floors met AND weighted avg ≥ 7.5? → ACCEPT
    ↓ (else)
paper-draft revises (max 3 attempts total)
    ↓
If still not accepted → flag for human review

Dimension	Weight	Hard floor
Novelty	30%	≥ 6.5
Rigor	25%	≥ 7.0
Literature coverage	20%	≥ 6.5
Clarity	15%	≥ 6.0
Impact	10%	≥ 6.0

Accept requires weighted avg ≥ 7.5 and all five floors met.

Journal Templates & Profiles

Templates enforce correct structure, section ordering, word limits, and formatting. paper-covert additionally loads a YAML profile that drives LaTeX conversion.

Markdown templates (`templates/`)

Category	Journals
`geoscience/`	Nature Geoscience, Geophysical Research Letters
`remote_sensing/`	Remote Sensing of Environment, IEEE TGRS, ISPRS JPRS
`giscience/`	IJGIS, Transactions in GIS, Annals of AAG

Submission profiles (`skills/paper-covert/profiles/`)

aag_annals.yaml, generic.yaml, ieee_tgrs.yaml, ijgis.yaml, isprs_jprs.yaml, rse.yaml, tgis.yaml.

MCP Servers

Declared in .mcp.json. Setup notes in mcp/README_MCP.md.

Server	Purpose
`filesystem`	Read/write local files and datasets
`fetch`	Fetch web content (papers, data portals, journal pages)
`geo_mcp`	Spatial data: GADM, OSM Overpass, Census ACS, GEE (`mcp/geo_mcp_server.py`)
`arxiv_mcp`	ArXiv search, paper fetch, abstract parsing
`github`	GitHub repo reading and code management
`brave_search`	Web search for literature, datasets, documentation

Key Output Files

File	Written by
`output/LIT_REVIEW_REPORT.md`	`lit-review`
`output/IDEA_REPORT.md` / `NOVELTY_REPORT.md` / `IDEA_REVIEW_REPORT.md`	`idea-discovery-pipeline`
`output/refine-logs/FINAL_PROPOSAL.md` / `REFINE_REPORT.md`	`refine-research`
`output/refine-logs/EXPERIMENT_PLAN.md` / `output/EXPERIMENT_TRACKER.md`	`experiment-design`
`output/experiment/EXPERIMENT_RESULT.md` / `EXPERIMENT_LOG.md`	`deploy-experiment`
`output/experiment/data/` / `figures/` / `scripts/`	`deploy-experiment`, `spatial-analysis`
`output/AUTO_REVIEW_REPORT.md` / `REVIEW_STATE.json` / `review-rounds/`	`auto-review-loop`
`output/METHOD_DESCRIPTION.md`	`auto-review-loop`
`output/NARRATIVE_REPORT.md`	`generate-report`
`output/PAPER_PLAN.md`	`paper-plan`
`output/figures/`	`paper-figure-generate`
`output/manuscript/`	`paper-draft`, `paper-review-loop`
`output/papers/`	`paper-covert`
`output/reports/submit_check_*.md`	`submit-check`
`data/DATA_MANIFEST.md`, `data/raw/`	`data-download`
`output/PROJ_NOTES.md`	all skills (append-only, compact log)
`output/TELEMETRY.jsonl` (per-session) and `output/TELEMETRY_STAGES.jsonl` (per-skill)	`tools/telemetry_aggregate.py` (run by Stop hook)
`output/CONTRACT_VIOLATION.md`	any skill that detects a downgraded success criterion or other contract violation
`memory/MEMORY.md`, `handoff.json`	Stop hook

Project Structure

night_owl_research_agent/
├── CLAUDE.md                        ← Dashboard and project conventions
├── README.md                        ← This file
├── design_principle.md              ← Skill-level design principles (export → Excel via tools/)
├── design_principle_agents.md       ← Sub-agent design principles
├── settings.json                    ← Claude Code hooks, permissions, env vars
├── .mcp.json                        ← MCP server declarations
│
├── .claude/
│   ├── commands/
│   │   └── launcher.md              ← /launcher (only installed slash command)
│   └── agents/                      ← Specialist sub-agent definitions (9 total)
│       ├── orchestrator.md
│       ├── literature-scout.md
│       ├── synthesis-analyst.md
│       ├── gap-finder.md
│       ├── hypothesis-generator.md
│       ├── geo-specialist.md
│       ├── paper-writer.md
│       ├── peer-reviewer.md
│       └── citation-manager.md
│
├── skills/                          ← 22 workflow skills + knowledge/
│   ├── full-pipeline/SKILL.md
│   ├── lit-review/SKILL.md
│   ├── idea-discovery-pipeline/SKILL.md
│   ├── generate-idea/SKILL.md
│   ├── novelty-check/SKILL.md
│   ├── idea-review/SKILL.md
│   ├── refine-research/SKILL.md
│   ├── experiment-design/SKILL.md
│   ├── experiment-design-pipeline/SKILL.md
│   ├── deploy-experiment/SKILL.md
│   ├── data-download/SKILL.md
│   ├── spatial-analysis/SKILL.md
│   ├── auto-review-loop/SKILL.md
│   ├── generate-report/{SKILL.md, templates/}
│   ├── paper-writing-pipeline/SKILL.md
│   ├── paper-plan/SKILL.md
│   ├── paper-figure-generate/{SKILL.md, templates/}
│   ├── paper-draft/{SKILL.md, templates/}
│   ├── paper-review-loop/{SKILL.md, templates/}
│   ├── paper-covert/{SKILL.md, profiles/, templates/}
│   ├── submit-check/SKILL.md
│   ├── training-check/SKILL.md
│   └── knowledge/                   ← Domain reference files
│
├── tools/                           ← CLI utilities (called by skills + harness)
│   ├── arxiv_fetch.py
│   ├── semantic_scholar_fetch.py
│   ├── convert_skills_to_llm_chat.py
│   ├── export_design_principle_table.py        ← exports design_principle.md tables to Excel
│   ├── export_agent_design_principle_table.py  ← exports design_principle_agents.md tables
│   ├── telemetry_stage_marker.py               ← called by skill_marker hook (per-skill timing)
│   └── telemetry_aggregate.py                  ← called by Stop hook (session/stage telemetry)
│
├── configs/
│   └── default.yaml                 ← Scoring weights, domain keywords
│
├── templates/                       ← Project + paper templates
│   ├── EXPERIMENT_LOG_TEMPLATE.md
│   ├── EXPERIMENT_PLAN_TEMPLATE.md
│   ├── FINDINGS_TEMPLATE.md
│   ├── HANDOFF_TEMPLATE.json
│   ├── IDEA_CANDIDATES_TEMPLATE.md
│   ├── PAPER_PLAN_TEMPLATE.md
│   ├── RESEARCH_CONTRACT_TEMPLATE.md
│   ├── RESEARCH_PLAN_TEMPLATE.md
│   ├── REVIEW_STATE_TEMPLATE.json
│   ├── geoscience/ (nature_geoscience, grl_template)
│   ├── remote_sensing/ (ieee_tgrs, isprs_jprs, remote_sensing_env)
│   └── giscience/ (ijgis, transactions_gis, annals_aag)
│
├── harness/
│   ├── hooks/ (pre_tool_use, post_tool_use, skill_marker, stop_hook, notification)
│   └── prompts/system_geo.md
│
├── mcp/                             ← MCP server implementations
│   ├── geo_mcp_server.py
│   └── README_MCP.md
│
├── memory/MEMORY.md                 ← Persistent session memory
│
├── output/                          ← All generated outputs
│   ├── AUTO_REVIEW.md
│   ├── REVIEW_STATE.json
│   ├── ARCHITECTURE_DIAGRAM_PROMPTS.md
│   ├── papers/
│   ├── figures/
│   └── reports/
│
├── res/nora_architecture.png        ← Architecture diagram
│
└── archived/                        ← Retired skills and pre-skill Python modules

Contributing

Fork this repository.
Add skills in skills/<name>/SKILL.md.
Add journal templates in templates/ (plus a YAML profile in skills/paper-covert/profiles/ if needed).
Add domain knowledge in skills/knowledge/.

License

MIT License. See LICENSE for details.

Design Principles

Two living documents describe the rules NORA's skills and sub-agents follow. Treat them as the source of truth when you write or change a skill.

File	Scope
`design_principle.md`	Skill-level principles: anchored problem, smallest adequate mechanism, generator–evaluator separation, conditional geospatial checks, mandatory local-GPU check before any pilot or full experiment, human-checkpoint pattern for synthesis decisions.
`design_principle_agents.md`	Sub-agent principles for the 9 specialists in `.claude/agents/` (orchestrator, literature-scout, gap-finder, hypothesis-generator, geo-specialist, paper-writer, peer-reviewer, citation-manager, synthesis-analyst).

Both files can be exported to Excel for review or workshop use:

python tools/export_design_principle_table.py
python tools/export_agent_design_principle_table.py

Inspired By

NORA's design borrows ideas from several open-source projects. Credit and gratitude to their authors:

BZBarrett/superpowers — skill-pack patterns for extending Claude Code with composable Markdown workflows.
BZBarrett/get-shit-done — pragmatic harness patterns for getting long-running agentic work to actually finish.
wanshuiyin/Auto-claude-code-research-in-sleep — "research while you sleep" autonomous-loop concept that motivated NORA's overnight pipelines, handoff.json recovery, and adversarial review loop.
karpathy/autoresearch — generator–evaluator separation and the per-criterion floors + weighted-average scoring loop adapted into auto-review-loop and paper-review-loop.

If your project influenced NORA and is missing here, please open an issue and we will add it.

Citation

If you use NORA in your research, please cite the arXiv preprint:

Zhou, B., Wu, Q., Huang, X., Ning, H., Li, D., & Zhang, Z. (2026). NORA: Night Owl Research Agent — Autonomous AI Research for Geoscience, Remote Sensing, and GIScience. arXiv:2605.02092. https://arxiv.org/abs/2605.02092

@misc{zhou2026nora,
  title         = {NORA: Night Owl Research Agent --- Autonomous AI Research for Geoscience, Remote Sensing, and GIScience},
  author        = {Zhou, Bing and Wu, Qiusheng and Huang, Xiao and Ning, Huan and Li, Diya and Zhang, Ziyi},
  year          = {2026},
  eprint        = {2605.02092},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2605.02092},
  howpublished  = {\url{https://github.com/GRIND-Lab-Core/night_owl_research_agent}}
}

Add this server

Host a copy with our keys and logs.

Server facts

Tools

Questions

How do I install Night Owl Research Agent?

Does Night Owl Research Agent need an API key?

What transport does Night Owl Research Agent use?

When was Night Owl Research Agent last checked?

Documentation

NORA (Night Owl Research Agent)

Quick Start

Step 1 — Install Claude Code

Step 2 — Get NORA onto your machine

Step 3 — Install the skills into Claude Code

Step 4 — Start a research session

Step 5 — (Optional) Enable extras

Prerequisites summary

What It Does

Architecture

Skills

Workflow Skills

Domain Knowledge

Control Flags

Harness Engineering

Autoresearch Scoring Loop

Journal Templates & Profiles

Markdown templates (templates/)

Submission profiles (skills/paper-covert/profiles/)

MCP Servers

Key Output Files

Project Structure

Contributing

License

Design Principles

Inspired By

Citation

Add this server

Host a copy with our keys and logs.

Server facts

Tools

Questions

How do I install Night Owl Research Agent?

Does Night Owl Research Agent need an API key?

What transport does Night Owl Research Agent use?

When was Night Owl Research Agent last checked?

Documentation

NORA (Night Owl Research Agent)

Quick Start

Step 1 — Install Claude Code

Step 2 — Get NORA onto your machine

Step 3 — Install the skills into Claude Code

Step 4 — Start a research session

Step 5 — (Optional) Enable extras

Prerequisites summary

What It Does

Architecture

Skills

Workflow Skills

Domain Knowledge

Control Flags

Harness Engineering

Autoresearch Scoring Loop

Journal Templates & Profiles

Markdown templates (templates/)

Submission profiles (skills/paper-covert/profiles/)

MCP Servers

Key Output Files

Project Structure

Contributing

License

Design Principles

Inspired By

Citation

Markdown templates (`templates/`)

Submission profiles (`skills/paper-covert/profiles/`)

Markdown templates (`templates/`)

Submission profiles (`skills/paper-covert/profiles/`)