Skip to content
ai-standards

Test ownership

In any pipeline that has both a Developer agent and a Tester agent, two contracts can fight each other:

  • The Developer’s Definition-of-Done verification gate says “every DoD checkbox must have a verified artefact, or the handoff is blocked”.
  • The Tester’s Role says “writes and executes all tests after implementation and review are complete”.

If the task’s DoD lists test rows under the Developer’s section (e.g. “Composable unit tests written”), both contracts apply at once: the Developer is forced to write the test to clear the gate, and the Tester writes their own version later. The result is the duplication you would expect — two agents producing partially overlapping tests, billed twice in Opus and Sonnet tokens.

The framework resolves this with a partitioned DoD and an explicit ⚠️ Tester scope mark.

The four-section task DoD

templates/feature-task-template.md declares four DoD subsections:

SubsectionOwnerExample rows
### Backend (Developer scope)Backend DeveloperCode passes PHPStan L9, OpenAPI annotation present, controller-level validation in place, LoggingMiddleware wired
### Frontend (Developer scope)Frontend DeveloperCode passes ESLint + vue-tsc, follows frontend.md
### Tester scopeTesterUnit / integration / composable / page tests written, all tests pass, Playwright captures for visual + interactive items
### SharedReviewer + DoD-checker + final phaseReviewer-approved, no security vulns, spec updated via /update-specs

Test artefacts (every flavour: PHPUnit, Vitest, Playwright captures) live exclusively under ### Tester scope. The Spec Analyzer is required to place them there when authoring the task file.

The ⚠️ Tester scope mark

Both Developer agents declare a four-mark ## DoD coverage table:

MarkMeaningSet by
✓Artefact verified on disk via grep/ls/ReadDeveloper
✗Artefact claimed but missing — handoff is blockedDeveloper
⚠️ Tester scopeRow lives under ### Tester scope of the task DoD; deferred to TesterDeveloper (mandatory for every test row, never ✓)
⚠️ (other)Verifiable only manually (e.g. multi-service smoke no automated tool can drive)Developer

The Tester later re-marks every ⚠️ Tester scope row in their own ## DoD coverage with ✓ / ✗ / ⚠️ after writing the artefact.

Routing on ⚠️ Tester scope

The DoD-checker is the gate between the Developer and the Reviewer. Per agents/dod-checker-agent.md, it carries ⚠️ Tester scope rows forward to APPROVED without verification:

  • Verifying them as if they were ✓ would either duplicate the Tester’s work (when the test happens to already exist from a prior iteration) or false-fail (when it does not exist YET because the Tester has not run).
  • The contract closure happens later, in the Tester’s ## DoD coverage section.
  • The DoD-checker does cross-check that the ⚠️ Tester scope mark was applied to a row that genuinely lives under ### Tester scope. Misuse on an architecture / wiring / config row is downgraded to ✗ with a citation.

Why this saves tokens (without sacrificing quality)

Empirical band measured against the previous shape (Dev writes tests to clear DoD, Tester rewrites or augments them):

PhaseBeforeAfterSaving
Backend Developer iter 1130-180k Opus~110-160k Opus~15-25k Opus
Frontend Developer iter 1160-200k Opus~140-180k Opus~15-25k Opus
Tester36-110k Sonnetunchanged0

Quality is preserved because the work is not removed — it is consolidated under a single specialised agent (the Tester) that already has the test-design context loaded (spec rules, fixtures, Playwright tooling, the tester-bundle.md). Removing the Developer’s parallel attempt eliminates a class of problems where the Developer wrote a test that the Tester then had to delete or refactor.

What the agents declare

  • agents/backend-developer-agent.md — verification gate step 2: “Rows under ### Tester scope are NOT yours to satisfy […] Mark every ### Tester scope row as ⚠️ Tester scope and skip the artefact verification”.
  • agents/frontend-developer-agent.md — same clause; adds Playwright as a Tester-scope artefact.
  • agents/tester-agent.md — Role section declares the Test-ownership contract; output mandates a ## DoD coverage re-marking every ⚠️ Tester scope row.
  • agents/dod-checker-agent.md — Work Loop step 3 carries ⚠️ Tester scope forward without verification; step 5 rejects ✓ marks that cite a test path.
  • agents/spec-analyzer-agent.md — Responsibilities require placing every test row under ### Tester scope.

Enforcement

  • Smoke check 25 anchors the contract in five places (template + four agents). If any of them drops the marker, CI fails on the smoke check.
  • The dynamic smoke (make smoke-dynamic) exercises the live orchestrator against fixture projects and observes the DoD-checker’s behaviour on ⚠️ Tester scope rows.