Feat/deploy ai agent prompt

[akmadan]

Summary

• Adds a deploy-ai-agent prompt template (src/prompts/deploy-ai-agent.ts) that walks an end-to-end workflow for shipping an AI agent from an IDE to a cloud agent runtime (Google Agent Runtime or AWS AgentCore) via Harness CD.
• The workflow (Phases A–G): get code into a Harness-reachable repo → confirm target platform → build & push the agent image (CI) → ensure connectors → create the AiAgent service → create environment + infrastructure → add the CD deploy stage and run it, with a failure/retry loop.
• Registers the prompt in src/prompts/index.ts and adds tests/prompts/deploy-ai-agent.test.ts (9 tests). README prompt count regenerated 34 → 35.

Why

Harness CD recently shipped AI agent deployment (service type AiAgent; infra kinds GoogleAgentRuntime / AwsAgentCore). This prompt brings that flow to the MCP server so an IDE coding agent can drive "ship this agent" reliably.

The prompt encodes the non-obvious contract:

• The agent image is a URI string on the service source (platform.spec.source.spec.image), not an artifact source.
• A build step is mandatory (only source type is a pre-built container).
• The built image is chained into the service via a shared <+pipeline.sequenceId> tag.
• AwsAgentCore requires executionRoleArn.
• Deploy/Shift/Rollback steps read the image from the service at runtime — never from the step.

It embeds the CI BuildAndPushGAR / BuildAndPushECR steps and all six CD steps (Deploy/Shift/Rollback per platform).

Notes

• Pure prompt addition — no new tools, no registry/schema changes.
• harness_schema in the prompt relies on live NG /yaml-schema fallback to surface the AiAgent types on FF-enabled accounts (no snapshot re-vendoring included).

Test plan

• pnpm typecheck — clean
• pnpm standards:check — 77/77
• pnpm docs:check — README up to date
• pnpm test tests/prompts/ — 72/72 (incl. 9 new)
• Confirm NG /yaml-schema surfaces AiAgent under the FF before merge

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[cursor]

Review summary

Solid addition — the phased A→G workflow mirrors build-deploy-app well, and the prompt correctly encodes the non-obvious AiAgent CD contract (lowercase container source discriminator, image URI on the service — not artifact source — <+pipeline.sequenceId> tag chaining, executionRoleArn for AwsAgentCore). The second commit fixing the container discriminator + the regression test show good schema diligence.

I ran the new test file locally: 9/9 passing.

Verdict: Approve with suggestions — none are blockers, but a few gaps could cause agent confusion during real "ship from IDE" runs.

Strengths

• Clear phase ordering with confirmation gates before writes
• Platform-specific YAML for both GoogleAgentRuntime and AwsAgentCore (build, service, infra, deploy/shift/rollback)
• registerAllPrompts wiring test guards against accidental removal
• README prompt count bump (34 → 35) satisfies docs:check

Suggestions (see inline)

1. Add Dockerfile / container-build prerequisites in Phase C (similar to build-deploy-app Phase 1)
2. Include a minimal full-pipeline skeleton in Phase G (codebase + CI stage + CD stage wiring), not just deploy steps
3. Document the hardcoded deploy stage identifier used in the revisionId expression
4. Consider adding org_id alongside projectId for scope symmetry
5. Clarify in the description that this deploys an AiAgent CD runtime — distinct from the create-agent prompt (Harness AI agent templates)
6. Add a row to the README MCP Prompts → DevOps table for discoverability (count-only update isn't enough for users browsing the table)
7. Open test-plan item: confirm NG /yaml-schema surfaces AiAgent on FF-enabled accounts before merge (entity schemas aren't vendored yet)

Happy to re-review after updates.