| name | mz-create-tasks |
| description | Decompose a PRD and TechSpec into a structured tasks.json task list for monozukuri's code phase. Use when the orchestrator routes the tasks phase. Do not use for PRD or TechSpec generation. |
| argument-hint | [feature-id] |
| version | 1.0.0 |
You are executing the tasks phase of monozukuri's autonomous feature loop.
Inputs available
- PRD at
$MONOZUKURI_WORKTREE/tasks/prd-$MONOZUKURI_FEATURE_ID/prd.md
- TechSpec at
$MONOZUKURI_WORKTREE/tasks/prd-$MONOZUKURI_FEATURE_ID/techspec.md
- Project conventions — from
CLAUDE.md, AGENTS.md, or the monozukuri learning store
- The prompt template at
references/tasks-template.md — describes the expected output format
- The validation rules at
references/tasks-validation.md
- The output schema at
references/tasks-schema.md (canonical JSON schema: schemas/tasks.schema.json)
Output contract
Write the task list to $MONOZUKURI_WORKTREE/tasks/prd-$MONOZUKURI_FEATURE_ID/tasks.json as a JSON array matching the schema in references/tasks-schema.md.
The file MUST satisfy references/tasks-validation.md. The validator runs immediately after; failures trigger a reprompt.
Hard rules
- Every task must be completable in ≤ 60 minutes of agent time.
- Every task must touch ≤ 5 files.
- Every task must have at least one verifiable acceptance criterion.
- Every FR and NFR from the PRD must appear in at least one task.
- Return ONLY the JSON array — no markdown, no commentary, no code fences.
- In autonomous mode, never block. Decompose at the granularity the TechSpec implies.
Workflow
- Read the PRD and TechSpec at the input paths.
- Read
references/tasks-template.md for the output format.
- Read
references/tasks-schema.md for the JSON schema.
- Read
references/tasks-validation.md.
- Identify every component, file change, and test from the TechSpec.
- Group into tasks satisfying the time and file-count constraints.
- Write the task list as valid JSON to
$MONOZUKURI_WORKTREE/tasks/prd-$MONOZUKURI_FEATURE_ID/tasks.json.
Workflow memory
Before starting, read $MONOZUKURI_MEMORY_DIR/MEMORY.md (if it exists) for shared context from earlier planning phases. After writing tasks.json, update $MONOZUKURI_TASK_MEMORY with the decomposition rationale and any tasks flagged as risky. If $MONOZUKURI_NEEDS_COMPACTION is non-empty and not none, run the mz-workflow-memory skill to compact memory files before continuing.
