version: "1.0.0" name: build-test description: Run the project's build / typecheck / lint / test commands and emit the build.passing + tests.passing signals devloop convergence reads. od: scenario: code-migration mode: critique

Build test

Spec §20.2 / §22.4: the "subjective critique-theater panel" is not enough on its own — a code-migration / tune-collab run must prove the build still passes. This atom shells out to the project's declared test commands and emits structured signals the devloop's until evaluator reads.

Inputs

• The active target stack (recorded in code/index.json +

plan.md's targetStack block).

• The user's optional override (`od plugin run --input

testCommand='pnpm test'`).

Default commands by framework

Output

project-cwd/
└── critique/
    ├── build-test.json   # { build: 'passing' | 'failing', tests: 'passing' | 'failing' | 'skipped', durationMs, commandsRun: [...], failures: [...] }
    └── build-test.log    # raw stdout / stderr (truncated to the runner's logBudgetBytes limit, default 1 MiB)

The atom emits two signals the devloop reads:

• build.passing: boolean — the typecheck command exited 0.
• tests.passing: boolean — the test command exited 0.

Plus the legacy critique.score so existing pipelines that read the score keep working: 5 when both pass, 3 when only build passes, 1 when both fail.

Convergence

Pipelines wire the atom into a devloop:

{
  "id": "verify",
  "atoms": ["patch-edit", "build-test"],
  "repeat": true,
  "until": "(build.passing && tests.passing) || iterations >= 8"
}

Anti-patterns the prompt fragment forbids

• Running tests on a partially-edited working tree (the atom

always runs after patch-edit, and only when plan.steps's current step is in completed state).

• Suppressing test failures by changing test files; that violates

the rewrite-plan ownership rule.

• Setting tests: 'skipped' without a non-empty reason field.

Status

Implemented by the daemon runner in apps/daemon/src/plugins/atoms/build-test.ts. It executes the configured build and test commands, bounds runtime and captured logs, and persists the report and log outputs above.