Skip to main content
Stage 2 of the agentic development lifecycle (ADLC) · run with /build.

​
What goes in, what you type, what comes out

Build takes a validated agentspec.yaml and implements it into the target you chose — a framework or a coding agent. It works test-first: it lays down the checks the spec implies, then builds until they pass.

​
Two agent steps

/build runs in two agent steps. Before each one, Helix names the step and the agent it hands it to.
1

Plan + Implement

An engineer agent reads the spec, the target’s documentation and your repository. It writes a plan into the build report before it touches anything, then builds it: tests first, then the code until they pass. The same agent runs the gates: lint, typecheck, build, tests and a check that every requirement in the spec is covered.For an agent deployed by API, with no repository to build in, the engineer writes a short plan and edits the agent’s definition (prompt, config, tools) instead. It prepares the change but never deploys it.
2

Review

A reviewer agent checks the result against the spec and against the plan. It is read-only: it reads and reports, it never edits. It ends with one verdict: proceed, send the work back with the findings to fix, or stop and ask you.
When the reviewer sends the work back, the engineer fixes the findings it cites and the reviewer reviews again. You get the finished agent and a build report: the plan, what was built, the test results, every place the result differs from the plan, and the reviewer’s verdict.

​
Run it

Point Helix at the spec and the target:
Other ways people ask:
  • Implement the Lead Qualification agent for Codex — enrich each lead, score it, and route the hot ones to sales — and stop at a working agent I can run.
  • Add the escalation rule so the Refund Processing agent hands disputes over $500 to a human instead of auto-approving.
Build implements the spec, runs its checks, and stops when you have a working agent.

​
Keeping spec and code in sync

Code drifts from the spec as you iterate. When that happens, type /sync-spec or ask Helix to resync the spec. It reconciles agentspec.yaml against what the implementation actually does, so the definition and the code stay in agreement. Any change to the spec is shown to you before it’s written. It also works on an agent that never had a spec: point it at the existing code and it writes one from what the code does.

​
What you get

  • A working agent that implements your spec.
  • A spec and an implementation that agree.

Next: Evaluate

Score the agent against your real traces.