Shrutam CCA-F Crash › Domain 3 › d3-l09-claude-code-cicd Hinglish English →

CI/CD with Claude Code — Headless Mode + Outputs

Domain 3 · 20% ~12 min Hinglish narration

Audio-only (commute / mobile data)

Same Saavi narration, smaller file. Opus 48k preferred — auto-selected by your browser.

Scenario anchor

Aaranya IT ke ek BFSI client — ek major private bank — ka scenario sochiye: 400+ microservices, har sprint mein 200 pull requests, aur compliance team ka pressure ki har code change ka audit trail hona chahiye. Manual review bottleneck ban gaya hai, aur aap ke Jenkins pipeline mein Claude Code ko integrate karna hai — but bina interactive UI ke, fully headless, structured JSON output ke saath taaki downstream tools parse kar sakein. Exactly yahi lesson resolve karta hai: how to run Claude Code in non-interactive CI mode, control its outputs, and trust the contract.

Key Takeaways

Claude Code headless mode is invoked with the --headless flag plus a non-interactive prompt; it exits cleanly with a machine-readable output rather than waiting for human input, making it safe to embed inside Jenkins, GitHub Actions, or any CI orchestrator.
Always declare a structured output contract — JSON schema or equivalent — and validate Claude Code's response against it before any downstream pipeline step consumes it; treat this exactly as you would validate an API response against an OpenAPI spec.
CI authentication uses API tokens injected as environment variables (ANTHROPIC_API_KEY); never bake credentials into Dockerfiles or pipeline YAML, and scope token permissions at the org level to contain blast radius if a pipeline credential is rotated or leaked.
Memory anchor: schema-validation-pipeline — Claude Code in CI is a service call with a schema contract, not a chat session; if the output does not validate, fail the pipeline stage fast, just as a schema-validation gate rejects a malformed event before it poisons the event bus.

← Previous lesson Same lesson in English Next lesson → Back to overview