Your first Claude API call. A one-shot agent reads a CI failure log and returns structured JSON with a diagnosis — no loops, no tools, no multi-step reasoning. Just a system prompt, an API call, and clean structured output. This is the pattern every later module builds on.

Run this before anything else. It checks Python version, SDK installation, API key, and output directory.

python module1/verify_setup.py

Expected output:

✅ Python 3.11.x
✅ anthropic 0.x.x installed
✅ ANTHROPIC_API_KEY found
✅ output/ directory writable
Environment ready.

If a check fails, the message tells you exactly what to fix:

• Missing SDK → pip install anthropic
• Missing key → export ANTHROPIC_API_KEY=sk-ant-...

# See the expected output shape without an API key:
python module1/hello_claude.py --mock

Expected output (mock mode):

{
  "summary": "3 test assertions failed in auth.test.js with memory climbing to 87%.",
  "likely_cause": "Uncleaned test fixtures retaining references between test cases, causing heap growth.",
  "next_step": "Add explicit cleanup in the afterEach hook and reduce fixture dataset size."
}

# Live call against Claude:
ANTHROPIC_API_KEY=sk-... python module1/hello_claude.py

# Saves full output to output/output_module1.json:
ANTHROPIC_API_KEY=sk-... python module1/agent.py

Key Takeaway:

• The system prompt is the program — not the code.
• Claude returns the same three keys every time because the system prompt specifies the exact JSON schema.
• This predictability is what makes agents composable.

Open hello_claude.py. You will find SYSTEM_PROMPT = "" and a set of TODO comments describing exactly what the prompt must do. Your task: write the prompt that instructs Claude to return summary, likely_cause, next_step, and confidence as valid JSON.

The run_api_mode() function and the ask() call are already wired — you only need to write the prompt. Run --mock first to confirm your environment works, then run live once your prompt is in place.

Attempt your own implementation before reading solutions/solution.py.

Workflow file: .github/workflows/module1-hello-agent.yml

The workflow runs automatically when you push any change inside the module1/ or shared/ folders. You can also trigger it manually: Actions tab → "Module 1 — Hello Agent" → Run workflow.

After the run completes, open the workflow run and click the Artifacts section at the bottom to download module1-output and inspect the JSON output your agent produced.

Prerequisite: Add your API key as a repository secret named ANTHROPIC_API_KEY (Settings → Secrets and variables → Actions → New repository secret).

• verify_setup.py reports all four checks green
• hello_claude.py --mock prints valid JSON with no parse errors
• Live run returns all three keys: summary, likely_cause, next_step
• Output saved to output/output_module1.json
• GitHub Actions workflow completes and module1-output artifact is attached to the run