---
name: palatineai-colosseum-research
description: Agent-readable Colosseum project research and opportunity evaluation layer. Use when researching hackathon projects, token candidates, founder quality, branding, or early opportunities.
version: 0.2.0
---

# PalatineAI Colosseum Research

Use this skill when researching hackathon projects, token candidates, founder quality, branding, or early opportunities from the Colosseum ecosystem.

PalatineAI turns a structured research database into a ranked intelligence layer agents can query before starting from zero: 2,857 projects researched, 1,680 fully scored across 12 signals, with the **top 120 published** as a machine-readable feed.

## Data source

- Live dashboard: https://palatineai.site/
- Structured data (JSON): https://palatineai.site/projects.json
- This skill file: https://palatineai.site/skill.md

### Quick start

```bash
# Top 10 projects by total opportunity score
curl -fsSL https://palatineai.site/projects.json | jq '.projects[:10] | map({rank, project, category, total})'

# Strongest branding signals
curl -fsSL https://palatineai.site/projects.json | jq '[.projects[] | {project, branding: .scores.branding, total}] | sort_by(-.branding)[:10]'

# Token-watch candidates: no token yet, strong social + brand momentum
curl -fsSL https://palatineai.site/projects.json | jq '[.projects[] | select(.scores.noToken >= 8) | {project, momentum: (.scores.socialActivity + .scores.branding + .scores.ideaQuality), x}] | sort_by(-.momentum)[:10]'
```

## Dataset shape

Top-level object:

| Field | Meaning |
|---|---|
| `stats` | Database-wide counts: `totalProjects` (2,857 researched), `scoredProjects` (1,680 scored), `manualReview` (1,177 still queued for manual scoring in the full DB), `categories`, `countries`, `noTokenSignals`, `publishedProjects`, `updated` |
| `projects` | Array of the top 120 scored projects, ordered by `rank` |

Each project:

| Field | Meaning |
|---|---|
| `project` | Project name |
| `description` | Team-submitted summary |
| `category` | One of 33 categories (DeFi, Consumer Apps, AI Platforms / Agents, DePIN, …) |
| `country`, `team`, `website`, `x` | Origin country, founder handles, official site, X profile |
| `scores` | 12 signals, each 0–10 (see below) |
| `total` | Sum of the 12 signals — max 120 |
| `rank` | Position by `total` within all scored projects |
| `tokenStatus` | `"No token signal"` or `"Possible token/funding signal"` |
| `manualReview` | Always `false` in the published feed (unscored projects are not published) |

### The 12 scored signals (`scores.*`)

`githubActivity`, `githubQuality`, `branding`, `country`, `teamExperience`, `ideaUniqueness`, `ideaQuality`, `idea`, `socialActivity`, `noToken` (10 = no token exists yet — an opportunity signal), `funding` (10 = not yet heavily funded), `additional` (manual bonus).

## What agents can query

- Top ranked projects by total opportunity score
- Token launch candidates (`noToken >= 8`, sorted by social + branding + idea momentum)
- Best branding scores
- Strongest founder/team signals (`teamExperience`)
- Most unique ideas (`ideaUniqueness + ideaQuality`)
- Possible token/funding signals (`tokenStatus != "No token signal"`)

Note: the published feed contains only fully scored projects. The 1,177 projects still awaiting manual review exist in the full database (see `stats.manualReview`) but are not in `projects.json`.

## Suggested workflow

1. Fetch https://palatineai.site/projects.json
2. Filter by category, token status, branding, founder strength, or idea uniqueness.
3. Shortlist projects with high `total` and clear distribution/token catalysts.
4. Deep-research the shortlist (website, X activity, GitHub) before action.
5. Log decisions as `TRACK`, `CONTACT`, `BUY/WATCH`, or `SKIP`.

## Example agent queries

- "Show the top 10 Colosseum projects by total opportunity score."
- "Find projects with no token yet and strong branding/founder signals."
- "Which teams look most likely to ship after the hackathon?"
- "Which categories have the highest average scores?"

## Install

Claude Code / Claude agents:

```bash
mkdir -p ~/.claude/skills/palatineai && curl -fsSL https://palatineai.site/skill.md -o ~/.claude/skills/palatineai/SKILL.md
```

Any other agent: fetch this file and `projects.json`; both are static, no auth required.