Quality gates and integration

Prepare your repository for the remaining HADR activities. You built the report and incident stores, then built the agent loop. Now add automated quality checks, merge open changes, and record a baseline run.

Do all work on this page in your own repository. Agree on the quality tools with your team. Ask teammates to review the changes that you merge.

Activity: Quality checks

Add code quality checks that run after repeated AI edits. Run them from a pre-commit hook. These checks prevent a gradual reduction in code quality.

Use these standard Python tools:

1. ruff finds unused, inconsistent, or incorrect code.
2. pyright or mypy finds type errors.
3. pytest runs tests.

Add these three (and behave) to your repo and configure them as a pre-commit hook.

Discuss other useful checks with your team. Agree on a shared set of checks. Each team member must add these checks to their own repository.

The behave gate

The scenarios in starter/features/ were written before the code. Some scenarios will fail until you implement their features.

1. Add uv run behave as a pre-commit hook and run it. Each failed scenario identifies work that is not complete. You can bypass a local hook with git commit --no-verify, so the hook is not a quality gate.

2. Use GitHub CI as the quality gate. Run it for each pull request into master. Initially, run only the scenarios for completed features:

   - run: uv run behave --tags="@warmup"   # gate the store warm-up only
   

3. Add tags to the gate when you complete each stage. After Stage 1 passes, use --tags="@warmup or @stage1". Add @stage2 after you write the intake scenarios. Add @stage3 after you complete reconciliation. Keep completed stages in the gate to prevent later changes from breaking them.

4. Add a rule to CLAUDE.md: check the CI tag expression when a feature file changes or a stage is completed.

When you’re done

Post your list of pre-commit hooks to the Padlet. Review the lists from other teams and add useful checks to your list.

Activity: Merging

1. Merge all open changes into your own master. This includes the phase 1 and phase 2 loop branches and the pull request from the evaluation driver. A teammate must review each pull request before you merge it. Share the reviews across your team.

2. Use a clean clone or worktree. Run uv run behave --tags=@warmup for the stores. Run uv run behave features/agent_loop.feature for the loop. Then run the skeleton episode with uv run hadr-runner stage1-basic@londone --no-llm. Start the engine first with uv run https://dl.hadr.ocelliq.com/hadr-engine.py serve --port 8000. The skeleton saves nobody. Record this result as the baseline.

3. Run the live loop on some examples (uv run hadr-agent "what's the secret message?") and record the rounds and token total it prints.

4. Check the prompt cache on the live run. agent_loop sends each assistant turn back without changes. Thus, the system prompt, tool definitions, and prior turns stay the same across rounds. The kit adds the provider’s prompt_tokens_details.cached_tokens value to result.usage.cache_read. Add cache_read to the final output in main(). Confirm that it is not zero for a run with multiple rounds.

When you’re done

1. Post your run on the Padlet. Include the rounds, total tokens, and cache_read from step 3. A cache_read of zero can mean that code changes the assistant turn. Compare your code with code from a teammate who has a nonzero value.
2. Complete the integration. Merge all required pull requests. Confirm that CI passes on master. Install the pre-commit hook in the clone that you will use for the remaining activities.

Leave with

• A reproducible setup command and test command.
• Automated checks that maintain code quality.
• A merged agent loop on your own master: green LOOP-* and store scenarios, plus a live hadr-agent run.
• The group intake fixtures merged into your repository, ready for Stage 2: intake.