magic-task

name	magic-task
description	Workflow for orchestrating tasks and generating the implementation plan.
handoffs	[{"label":"Generate plan","workflow":"magic-task","prompt":"Generate or update the implementation plan and tasks based on ALL registered specifications."},{"label":"Execute tasks","workflow":"magic-run","prompt":"Proceed to task execution without interruption to maintain Zero-Prompt workflows (Rule C9).","condition":"tasks_generated"}]

Task Workflow

Triggers: "Generate tasks", "Create tasks", "Update tasks", "Sync tasks", "Create plan", "Generate plan", "Update plan".

Trigger: /magic-task [arg]. Arguments:

• (empty) — full planning across all workspaces
• {workspace} — scoped planning for a specific workspace
• "text" — guided planning with focus or instructions
• {workspace} "text" — scoped + guided planning

Examples: /magic-task, /magic-task engine, /magic-task "decompose phase-2 in more detail", /magic-task docs "only new specs".

Scope:

• Rules Parity: RULES.md version must be recorded in TASKS.md header.
• Registry First: every plan update must synchronize with INDEX.md. Orphaned specs are critical blockers.
• Plan generation, task decomposition, and execution orchestration. Execution is handled by magic-run. Specification authoring by magic-spec. Rule governance by magic-rule.
• Pipeline: magic-spec → magic-task → magic-run.
• Post-Run Entry (rules/MAGIC-md §5): when invoked after /magic-run for drift recovery, Pre-Planning Stabilization auto-fixes mechanical drift (Draft→Stable promotion, field normalization, phantom backlog moves); Pre-flight HALTs with a single /magic-spec recommendation only on substantive gaps requiring human design input.
• Finalization: after plan write, run node .magic/scripts/executor.js finalize --workflow=task and display output verbatim. Never auto-commit. See .magic/task.md §Finalization Protocol.

Full implementation: .magic/task.md. Read it before proceeding. Executor: node .magic/scripts/executor.js <script> for all automation. Anti-Hallucination Guard: do not invent ad-hoc scripts (.js, .sh, etc.) for internal engine operations. Magic SDD steps are evaluated cognitively unless an executor script is explicitly provided.