Deep Planning is a high-latency System 2 planning protocol for AI agents that abstracts complex system-design problems, interrogates them via a blank-context domain-shifting subagent, and outputs a metaphor-washed, self-contained execution blueprint.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install deep-planning-skill
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install deep-planning-skill using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Deep Planning?
The Deep Planning skill is a powerful System 2 planning routine designed to tackle complex architecture, algorithm, system-design, and research-planning challenges that typically cause ordinary linear planning to fail. By utilizing asymmetrical expertise, this protocol abstracts a problem into a domain-neutral form and spawns an isolated, blank-context Domain Questioner subagent. This subagent analyzes the problem through the lens of a homologous field (such as control systems, traffic engineering, or CNC process control) to expose overlooked mechanisms, which are then translated back into the original domain as robust, actionable solutions.
Integrating this protocol into your Openclaw Skills workflow ensures that AI agents can overcome stubborn bottlenecks, avoid local assumptions, and systematically evaluate standard, novel, or hybrid planning options. The end result is a metaphor-free, highly structured execution blueprint that a downstream agent can run autonomously without needing the preceding chat history.
Deep Planning Use Cases
- Designing novel system architectures, complex API routing mechanisms, or high-scale concurrent systems.
- Generating lateral research plans and deep-research handoffs with clear validation metrics.
- Troubleshooting long-horizon execution plans where standard ReAct agent cycles have repeatedly failed.
- Formulating robust algorithm designs requiring cross-disciplinary structural analogies (such as mapping network congestion to traffic engineering).
- Creating standalone execution blueprints with built-in recovery protocols for autonomous downstream workers.
How Deep Planning Works
- Activation Gate: The main agent evaluates whether the task justifies the overhead of an advanced Openclaw Skills routine.
- Standard Baseline Plan: The agent defines and scores an initial, ordinary original-domain solution.
- Structural Abstraction: The core problem is stripped of domain-specific nouns and translated into a domain-neutral topology.
- Domain-Native Problem Framing: The abstract problem is rewritten to fit a selected homologous field with mature real-world mechanisms.
- Spawn Blank-Context Domain Questioner: An isolated subagent (the Domain Questioner) is spawned with zero knowledge of software or the original domain to interrogate the framed scenario.
- Forced Translation: The main agent extracts domain-native insights returned by the subagent and translates them back into practical implementation structures.
- Quantitative Comparison & Hybridization: The standard baseline, novel domain-shifted, and hybrid plans are scored and compared.
- Metaphor Wash & Hardening: All metaphorical residue is purged, and the plan is hardened into a self-contained execution contract with milestoning, validation, and recovery steps.
Deep Planning Setup
To utilize the Deep Planning protocol within your Openclaw Skills workspace, configure your agent system payload and configure support for blank-context subagents. If your environment does not support multi-agent spawning, the skill will gracefully degrade to internal simulation mode.
Ensure your configuration is declared inside your agent's system profile or skill path:
# Initialize the skill inside your agent environment
openclaw skills register deep-planning --path ./skills/deep-planning.md
Configure the mandatory system payload for the blank-context Domain Questioner as defined in the protocol to guarantee proper isolation.
Deep Planning Data Schema & Taxonomy
The Deep Planning protocol structures its output using strict document schemas to maintain alignment across agent handoffs. When executing this Openclaw Skills routine, the agent generates or maintains the following data models and plan logs:
Document Structuring Matrix
| Document Component | Description | Format / Key Fields |
|---|---|---|
| Baseline Scorecard | Scoring of the original standard plan (1-5 scale) | Correctness, Simplicity, Observability, Reversibility |
| Translation Ledger | Mapping of lateral concepts to literal software assets | Domain-native mechanism, Original-domain equivalent, Failure modes |
| Execution Blueprint | Self-contained, metaphor-washed implementation plan | Objective, Assumptions, Control Flow, Validation Matrix |
| Progress & Decision Log | State tracker for downstream worker execution | Milestone, Action, Result, Decision, Rationale, Alternatives |
Validation Schema Table
Every blueprint must define a validation matrix to verify task completeness:
| Success Criterion | Validation Method | Expected Result | Failure Response |
|---|---|---|---|
| Concrete requirements mapped | Tests/Commands/Citations | Quantitative outputs | Local repair or escalation |
Deep Planning Advanced Features
- Asymmetric Context Isolation: Employs absolute blindness rules to prevent the spawned subagent from collapsing back into ordinary, biased software advice.
- In-Domain Patch Loop: Recursively refines and patches secondary bottlenecks inside the metaphorical domain before attempting any global redesign.
- Recursive Scope Narrowing: Enables targeted pivoting to narrower, specialized fields if a broader metaphor encounters a physical or logical boundary.
- Quantitative Hybridization Scorecard: Uses strict metrics to compare standard, novel, and hybrid solutions, penalizing unneeded cleverness to guarantee simplicity.
- Self-Contained Execution Contracts: Hardens final output blueprints to support zero-context execution by downstream coding and research agents.
SKILL.md
Loading
Related Openclaw Skills
Health Monitor
Health Monitor is an automated resource tracking and self-healing system tool designed to keep low-resource environments running smoothly.
DGT Wait Assistant
An AI agent skill to query and anonymously report practical driving test wait times across different provinces in Spain.
AI Startup MVP Factory
An end-to-end automated pipeline that translates a single product idea into a fully-functional, containerized production-ready application.
GitHub Projects v2 Skill
An AI agent skill designed to seamlessly manage GitHub Projects v2 backlogs, status updates, and issue lifecycles via the GitHub CLI.
Backup Tool
A robust and secure utility to backup and restore OpenClaw configurations, skills, memory, identities, and credentials in a single command.
Yorkshire Puppy Care
An AI-powered health and care assistant designed specifically to automate schedules, track health milestones, and analyze behaviors for Yorkshire Terrier puppies.
Featured*
