| name | build |
| description | Runs the shared build stage: TDD implementation from the plan document's task list and two-agent code review with autonomous triage. Use standalone with an existing plan.md/fix-plan.md, or composed as the build stage of /belt:feature-dev and /belt:bug-fix. QA runs as its own stage (/belt:qa) after build; integration happens at the orchestrator's integrate phase. |
| user-invocable | true |
| argument-hint | [--codex] |
build
Belt pipeline for the build stage. Structure, gates, and done criteria
live in pipeline.yml; this file defines how to execute each phase.
Entry check
A plan document must exist: docs/features/<topic>/plan.md with a
Tasks section (feature runs) or docs/features/<topic>/fix-plan.md
(bug runs). If neither exists, stop and ask the user. If
docs/features/<topic>/evidence.md does not exist, create it now with
the header # Evidence: <topic>.
Phase: execute
This phase has no invoke — execute these steps directly:
- Read the plan document's task list (Tasks in plan.md, or the task
list in fix-plan.md).
- For each unchecked task, dispatch ONE
belt-agent:implementer
subagent with a self-contained prompt containing: the task text, the
exact file paths, the test(s) to write first, the relevant design
constraints copied into the prompt (never "see the design doc"), and
the project's test command.
- After each subagent returns: run the project test suite yourself,
check the task's checkbox in the plan document, and commit.
- Tasks whose target files do not overlap MAY run in parallel;
overlapping tasks run serially.
- Append the execute entry to evidence.md (test + lint commands and
observed results).
Phase: code-review
The invoke is declared in pipeline.yml. In pipeline mode its triage
is autonomous: critical/high findings are fixed and committed, or
recorded as deferred with a reason. Append the code-review entry to
evidence.md including the deferred list.
Red flags
- Never start execute without the Entry check.
- Never forward the whole plan document to implementer subagents — copy
the relevant constraints into each prompt.
- Never skip the per-task test run after a subagent returns.
- Never let subagents write evidence.md — orchestrator only.
References
plugins/belt/skills/design/references/path-convention.md — naming rules (SSOT)plugins/belt-agent/references/authoring-principles.md — evidence entry format
