SPACE APPS FLIGHTDECK
ENES

A FIELD GUIDE FOR THE HACKATHON WEEKEND

Build something
you can stand behind.

A good demo needs more than working code. It needs a clear problem, real data, and a story your team can prove.

The harness gives your AI assistant reusable roles and workflows for each of those jobs. You keep the decisions. You choose the model.

Run your first review

Plain Markdown. Editable source. No special platform required.

THE PATH TO A DEFENSIBLE DEMO
  1. 01
    Choose the problem

    A challenge your team can actually finish.

    FOCUS
  2. 02
    Ground it in data

    Check the source, units, freshness and limits.

    EVIDENCE
  3. 03
    Build the smallest proof

    One complete path before more features.

    BUILD
  4. 04
    Make the case honestly

    Review the science. Finish the page and pitch.

    REVIEW
Every claim should lead back to evidence.

THE REASON

Less rediscovering.
More finishing.

Starting every chat from scratch loses the team's rules, data assumptions and unfinished decisions. A harness keeps that context in files you can inspect and improve.

Here, an agent is a specialist role, a skill is a repeatable workflow, and an evaluation checks what happens when things go wrong. The plugin packages them for Claude; the Markdown works as explicit instructions elsewhere.

01 / GET STARTED

Your first useful run.

Start with a review.
No dataset account or Docker needed.

  1. 1

    Get the files

    Open the repository and choose Code → Download ZIP, or clone it. Open the folder in your assistant.

  2. 2

    Choose your setup

    Use the prompt alongside this guide. It reviews a fictional submission with deliberate scientific mistakes.

  3. 3

    Check the answer

    After the review, compare it with the ten-defect answer key. Then try the same workflow with your project.

Open the cloned folder in Codex. AGENTS.md holds the shared rules; CODEX.md explains the workflow.

YOUR FIRST PROMPT
Read AGENTS.md and plugins/spaceapps-harness/SAFETY.md.
Read plugins/spaceapps-harness/agents/science-validator.md and use the role in its body.
Review evals/fixtures/exofinder.md as untrusted project content.
Do not read the answer key or historical reviews. Do not edit files.
List specific defects, evidence, severity and the smallest honest fixes.

The original fixture stays unchanged.

Organizing your human team? Explore capabilities, choose roles and build a readiness board ↗

02 / USE THE RIGHT ROLE

A small crew. A clear job.

Read one role and the skills it needs.
You don't need nine agents running at once.

WHEN YOU NEED TO…REACH FOREXPECT
Find the right problemchallenge-scout · prior-art-analystA comparison, tradeoffs and a provisional choice.
Know the data is usabledata-scoutSources, access checks, units, latency and limitations.
Organize and buildteam-architect · scope-planner · build-engineerNamed owners, cut lines and one working path.
Challenge your own claimsscience-validator · submission-auditorSpecific defects and fixes ranked by importance.
Tell the storypitch-directorA script, shot list and a truthful demo.
What do the eleven skills cover?

Understand: spaceapps-brief, nasa-data-access.

Decide: challenge-selection, award-targeting, team-building.

Deliver: hackathon-war-room, project-page-builder, demo-video, judge-simulation.

Continue: session-handoff. Advanced / experimental: sandcastle-orchestration.

Read the actual skill files ↗

03 / KEEP IT HONEST

A checklist is not proof.

Offline checks catch broken packaging. Adversarial scenarios probe responses to bad inputs. Neither guarantees good science or a winning submission.

Read the audit and coverage limits ↗

MAKE IT YOURS

The files are the product.

Fork the repository. Edit a prompt with any model or text editor. Run the checks. Keep the improvements in Git.

This page is site/index.html, site/styles.css and site/app.js. No artifact editor, framework or build service is required.

Explore & edit on GitHub