Skip to main content
Illustrated Roman scroll with data traces — the PalatineAI skill
The skill is a single markdown file at https://palatineai.site/skill.md. It contains the dataset URL, the schema, ready-made queries, and a suggested evaluation workflow — everything an agent needs to use PalatineAI without scraping or guessing. Current version: 0.2.0 (version in the file’s frontmatter). Skill name: palatineai-colosseum-research.

​
Claude Code

User-level (available in every project):
User-level install
mkdir -p ~/.claude/skills/palatineai && curl -fsSL https://palatineai.site/skill.md -o ~/.claude/skills/palatineai/SKILL.md
Project-level (committed with a repo, shared with collaborators):
Project-level install
mkdir -p .claude/skills/palatineai && curl -fsSL https://palatineai.site/skill.md -o .claude/skills/palatineai/SKILL.md
Verify: ask Claude “What are the top 5 Colosseum projects by total score?” — it should fetch projects.json and answer with ranked rows, not a web search.

​
Any other agent

The skill file is plain markdown with YAML frontmatter (name, description, version). Use whichever integration your framework has:
  • System prompt / context: fetch the file and inline it. It is ~4 KB.
  • Tool description: register a palatine_research tool whose description is the skill body and whose implementation is GET https://palatineai.site/projects.json.
  • RAG / knowledge base: index skill.md and projects.json together; the skill explains the dataset’s fields.
Generic fetch
curl -fsSL https://palatineai.site/skill.md

​
When the skill should trigger

The skill’s own description field declares its scope — use it for routing:
Use when researching hackathon projects, token candidates, founder quality, branding, or early opportunities.

​
Staying current

The skill and dataset update independently. Re-run the install command to refresh the skill; check version in its frontmatter for changes. Dataset freshness is stats.updated inside projects.json (see Agent access for conditional fetches with ETag).