By name: verify-a4 description: Verify adoption criterion A4 (Task Scope) in the current project. Reports fulfillment level 0–3. allowed-tools: Bash Read
Verify A4 — Task Scope
Criterion Definition
| Level | Description |
|---|---|
| 0 | Agents only answer questions or explain code; they do not make changes |
| 1 | Agents handle bounded edits: a single function, a bug fix, a single-file change |
| 2 | Agents implement complete features spanning multiple files, writing and running tests as part of the implementation |
| 3 | Agents implement entire user stories end-to-end, including cross-service changes, test suites, documentation, and migration scripts |
Evidence to Gather
- Look at agent-co-authored commits in recent git history and assess how many files each changed — this indicates the scope of tasks agents routinely handle.
- Read skill definitions in
.claude/skills/to understand the scope of tasks they describe (single-function fixes vs. multi-file features vs. story-level work). - Read
CLAUDE.mdorAGENTS.mdfor any guidance on expected task granularity. - Look for task tracking files (
TODO.md,BACKLOG.md,stories/) and assess the granularity of items in them.
Instructions
Gather the evidence described above and determine the fulfillment level for A4.
This criterion measures the largest scope of work that agents routinely handle end-to-end in this project. Look for evidence in git history (file counts per agent commit), skill definitions, and CLAUDE.md/AGENTS.md guidance.
Scoring guide:
- Level 0: No evidence of agents making code changes — no co-authored commits in git history, skills only describe Q&A or explanation tasks.
- Level 1: Agent-authored commits exist but are consistently small — 1–2 files changed, targeted bug fixes, single-function additions. Skill definitions describe bounded tasks.
- Level 2: Agent-authored commits regularly span multiple files (3+), include test files alongside implementation files, and/or skills describe full feature implementation as the expected scope.
- Level 3: Agent-authored commits span services or packages, include migrations, documentation, and comprehensive test suites. Skills or CLAUDE.md describe user-story-level implementation as the target scope. TODO/backlog items are story-level, not sub-task-level.
Report in exactly this format:
A4 — Task Scope
- Level: [0 / 1 / 2 / 3]
- Rationale: [one or two sentences citing the specific evidence]