Goal Decomposer: Turn One Fuzzy Outcome Into an Executable Task Graph
Forces the agent to break a vague goal into a dependency-ordered DAG with tools, success checks, time estimates, and a flagged critical path BEFORE it touches anything.
Act as the planning module of an autonomous agent. Do NOT execute anything yet. Decompose this goal into an executable task graph.
Goal: [GOAL]. Available tools/resources: [RESOURCES/TOOLS]. Deadline: [DEADLINE].
Think first: identify the final deliverable, then work backwards to the atomic tasks that produce it. Split anything that bundles two verbs into separate tasks. For each task define a measurable success check (how the agent KNOWS it is done) and its dependencies (the task IDs that must finish first).
Then: flag the critical path, name the single riskiest task and why, and list 3 assumptions that, if wrong, break the plan.
OUTPUT FORMAT: (1) a JSON array of task objects, each {"id","task","tool","depends_on":[],"success_check","est_minutes"}; (2) a one-line ordered critical path of IDs; (3) a 'Risks & Assumptions' bullet list. End with: 'Reply GO to execute task by task, or EDIT to revise the plan.' Do not act until I reply GO.Fill the highlighted [VARIABLES] with your details, then paste into your AI.
The whole vault is free
Every prompt — copy, customize the [BLANKS], and run it in ChatGPT, Claude & Gemini. No sign-up, no paywall.
More Agentic & System prompts
- The Universal Autonomous Runner: Plan-Act-Reflect Loop That Never Goes Rogue
- AI Employee Onboarding: Build a Reusable Operator System Prompt in One Shot
- Multi-Agent Orchestrator: Manager Splits the Job, Parallel Sub-Agents Crush It
- Reflexion Loop: Self-Critique Until Your Output Passes Its Own Rubric
- Grill-Me-First Agent: It Interviews You Before It Builds Anything
- Schema-First Action Agent: Closed-Enum Tool Calls That Never Emit Bad JSON