EliteaSkills

A Claude-Code-native toolkit for the ELITEA platform — 2 role-based agents and 4 progressive-disclosure skills covering the full ELITEA lifecycle: authoring pipeline YAML, wrapping APIs as toolkits, deploying agents, smoke-testing them, and scheduling recurring jobs.

Distributed as a Claude Code plugin (and via an npx installer for Cursor / Windsurf / GitHub Copilot).

What's in the box

agents/
├── elitea-builder/                 # builds ELITEA artifacts
└── elitea-integrator/              # connects external systems to ELITEA

skills/
├── elitea-platform/                # foundation: REST API, auth, IDs, MCP catalog
├── elitea-pipeline/                # pipeline YAML — schema, patterns, examples
├── elitea-toolkit/                 # toolkit types + real OpenAPI specs
└── elitea-testing/                 # predict, debug, schedule + nudge case study

Install — three ways

1. Claude Code plugin marketplace (recommended inside Claude Code)

/plugin marketplace add Bermudas/EliteaSkills
/plugin install elitea-skills@elitea-skills

Installs the full bundle (both agents + all four skills) in one shot. To install just a piece:

/plugin install elitea-builder@elitea-skills
/plugin install elitea-platform@elitea-skills

2. npx installer (works for Claude Code, Cursor, Windsurf, GitHub Copilot)

# everything
npx github:Bermudas/EliteaSkills init --all

# pick & choose
npx github:Bermudas/EliteaSkills init --agents elitea-builder
npx github:Bermudas/EliteaSkills init --skills elitea-platform,elitea-pipeline

# specific IDE only
npx github:Bermudas/EliteaSkills init --target claude --update

The installer copies agents/ and skills/ into the right IDE-specific dirs:

• Claude Code → .claude/agents/ and .claude/skills/
• Cursor → .cursor/agents/ and .cursor/skills/
• Windsurf → .windsurf/agents/ and .windsurf/skills/
• GitHub Copilot → .github/agents/<name>.agent.md (flat layout with model alias normalized)

3. Copy by hand

This repo is self-contained — agents/<name>/AGENT.md and skills/<name>/SKILL.md are ready to drop into your IDE's discovery directory.

After install — set up auth

Every consumer project needs an ELITEA Personal Access Token:

1. Open https://next.elitea.ai → Profile → API Tokens → generate one
2. In your project, create .env:

   cp .env.example .env
   $EDITOR .env
   # paste the token after ELITEA_TOKEN=

The token name is ELITEA_TOKEN in this repo. Older code or third-party docs may use other names — they all mean the same PAT:

Where	Env var name
Project .env	ELITEA_TOKEN ← canonical
skills/elitea-testing/scripts/update_agent.py	ELITEA_API_TOKEN (falls back to ELITEA_TOKEN)
GitHub Actions secrets	ELITEA_NEXT_API_KEY (matches existing workflow conventions)
Inside an ELITEA pipeline (runtime)	DO NOT read .env — use elitea_client.auth_token

What each agent does

elitea-builder 🛠️

For building artifacts ON the ELITEA platform: agents, pipelines, toolkits, credentials, links.

Triggers on: "create an elitea agent", "build a pipeline", "wrap an API as a toolkit", "deploy this YAML", "link toolkit X to agent Y", "create a credential".

elitea-integrator 🔌

For calling ELITEA from somewhere else — external Python/JS scripts, GitHub Actions, Power Automate, Teams bots, JIRA webhooks, scheduled jobs.

Triggers on: "integrate with elitea from", "schedule a recurring run", "test my deployed agent", "debug why this conversation is stuck".

What each skill contains

Skill	When it gets loaded	Key files
elitea-platform	Any ELITEA work that needs REST endpoint details, auth, ID rules, MCP tool catalog. Load first when in doubt.	references/api-reference.md, references/conventions.md, references/mcp-tools.md
elitea-pipeline	Authoring/debugging pipeline YAML.	references/yaml-schema.md, references/patterns.md, references/workflows.md, examples/*.yaml
elitea-toolkit	Wrapping APIs, MCP servers, and first-class integrations as ELITEA toolkits.	references/toolkit-types.md, references/openapi-toolkits.md, examples/*.{json,yaml}
elitea-testing	Running, predicting, debugging, scheduling.	references/test-patterns.md, references/nudge-case-study.md, scripts/update_agent.py

Worked example — the nudge pipeline

For a complete end-to-end story (build → deploy → debug → schedule), read skills/elitea-testing/references/nudge-case-study.md. It walks through how ConversationHealthAnalyzer was built: deterministic classifier instead of LLM, the false-positive "hung" bug discovered on the first apply run, the idempotency design, deployment to two projects, and the GitHub Actions cron that runs it every 15 minutes.

The actual artifact files in this repo:

• Pipeline YAML: skills/elitea-pipeline/examples/ConversationHealthAnalyzer.yaml
• GH Actions workflow template referenced from the case study

Local usage workflow

After installing into a project, common operations:

Push a local instruction .md to a deployed agent

python3 .claude/skills/elitea-testing/scripts/update_agent.py path/to/instruction.md           # dry-run
python3 .claude/skills/elitea-testing/scripts/update_agent.py path/to/instruction.md --apply   # apply

Smoke-test a deployed agent version

source .env
curl -X POST -H "Authorization: Bearer $ELITEA_TOKEN" -H "Content-Type: application/json" \
  -d '{"user_input":"hello"}' \
  "https://next.elitea.ai/api/v2/elitea_core/predict/prompt_lib/$PROJECT_ID/$VERSION_ID"

Full test-pattern recipes (conversation flow, send-and-poll, async, debugging checklist) live in skills/elitea-testing/references/test-patterns.md.

Repository layout (developing this package)

EliteaSkills/
├── agents/
│   ├── elitea-builder/AGENT.md
│   └── elitea-integrator/AGENT.md
├── skills/
│   ├── elitea-platform/
│   ├── elitea-pipeline/
│   ├── elitea-toolkit/
│   └── elitea-testing/
├── .claude-plugin/
│   ├── plugin.json                 # top-level Claude Code plugin manifest
│   └── marketplace.json            # per-agent + per-skill installables
├── .cursor-plugin/plugin.json      # Cursor manifest
├── bin/init.mjs                    # npx installer
├── package.json                    # npm metadata + bin alias
├── README.md                       # this file
├── CLAUDE.md                       # routing hints for Claude Code
├── .env.example
└── .gitignore

Environment

ELITEA runs at https://next.elitea.ai/ — this is the only environment. The legacy nexus.elitea.ai host has been retired; if you find references to it in old code or docs, update them to next.elitea.ai.

Contributing

If you build a new pipeline or toolkit worth keeping:

1. Drop the YAML/JSON into skills/elitea-pipeline/examples/ or skills/elitea-toolkit/examples/.
2. Add a one-line entry in that skill's SKILL.md example catalog.
3. If it embodies a non-obvious pattern (idempotency, async polling, error recovery), document it in the relevant references/*.md.
4. Bump version in .claude-plugin/plugin.json, .claude-plugin/marketplace.json, .cursor-plugin/plugin.json, and package.json.

License

MIT — see LICENSE.

Credits

Packaging pattern (plugin manifests, multi-target init.mjs installer, sdlc-skills style agent frontmatter) inspired by arozumenko/sdlc-skills.