setup - SKILL.md Agent Skill

name: setup description: Interactive guide to set up Helix locally for development or contribution user-invocable: true

Help the user set up Helix locally, step by step. Walk through each stage interactively, checking the environment before moving on.

Setup stages

1. Prerequisites check

Check the following are available in the shell, and tell the user what to install if any are missing:

python3 --version → must be 3.12+
docker --version and docker compose version → required for full-stack mode
git --version

If uv is available (uv --version), recommend it for dependency installation (faster). Otherwise fall back to pip.

2. Clone (if not already inside the repo)

If the current directory is not the helix-community repo:

git clone https://github.com/88hours/helix-community.git
cd helix-community

3. Install Python dependencies

With uv (recommended):

uv pip install -e ".[dev]"

With pip:

pip install -e ".[dev]"

4. Configure environment

cp .env.example .env

Walk the user through filling in .env. The required and optional variables are:

Variable	Required	Notes
`REDIS_URL`	Yes	e.g. `redis://localhost:6379` — covered by Docker Compose
`GITHUB_TOKEN`	Yes	GitHub PAT with `repo` scope
`HELIX_GITHUB_REPO`	Yes	The repo Helix will post fixes to, e.g. `owner/repo`
`ANTHROPIC_API_KEY`	If using Anthropic	Get one at https://console.anthropic.com/
`HELIX_PROVIDER`	No	`anthropic` (default) or `ollama`
`HELIX_MODEL`	No	e.g. `claude-sonnet-4-6` or `llama3.2`
`HELIX_OLLAMA_BASE_URL`	Ollama only	Default: `http://localhost:11434`
`SENTRY_WEBHOOK_SECRET`	One of these	Sentry client secret
`ROLLBAR_ACCESS_TOKEN`	One of these	Rollbar project read token
`SLACK_BOT_TOKEN`	Recommended	`xoxb-...` bot token
`SLACK_SIGNING_SECRET`	Recommended	Slack app signing secret
`SLACK_APPROVAL_CHANNEL`	Recommended	Channel name for notifications

Ask the user which LLM provider they want to use (Anthropic or Ollama) and help them set the right variables for their choice.

5. Start the stack

Option A — Full stack via Docker Compose (recommended for testing the full pipeline):

docker compose up --build

Option B — Individual agents without Docker (faster iteration on a single agent):

# Terminal 1 — webhook receiver
uvicorn agents.crash_handler.main:app --reload

# Terminal 2 — test case generator
python -m agents.qa.main

# Terminal 3 — fix suggestion generator
python -m agents.dev.main

# Terminal 4 — Slack notifier
python -m agents.notifier.main

Ask the user which mode they prefer and show only the relevant commands.

6. Verify with tests

Run the test suite to confirm everything is wired up correctly (no real API keys needed — all HTTP calls are mocked):

pytest

Expected: 18 test files, all passing. If any fail, help diagnose.

7. Point a webhook (optional)

If the user wants to test the full end-to-end flow, tell them to point their Sentry or Rollbar webhook at:

Sentry: http://localhost:8000/webhook/sentry
Rollbar: http://localhost:8000/webhook/rollbar

Tips to share

The agents talk via Redis Pub/Sub — no agent calls another directly. If something seems stuck, check docker compose logs -f <agent>.
ruff check . must pass before opening a PR. Run ruff check . --fix && ruff format . to auto-fix most issues.
Tests live in tests/ and mirror the source structure. Add a test file for any new module you create.
For questions: hello@88hours.io