Introduction
The core purpose of the writing-plans skill is to translate high-level product specifications into highly granular, actionable micro-task lists before writing any actual code. It solves a critical engineering pain point: when facing complex multi-step tasks, AI agents or junior developers often lose context, compromise architecture, or skip testing. By enforcing a “zero-context” assumption and Test-Driven Development (TDD), it ensures every code change is precise and verifiable.
Concept
Transform macro requirements into test-driven micro-plans. Assuming a “zero-context” executor forces every code change to be deliberate, isolated, and strictly verified, preventing architectural decay from the start.
Setup and Usage
There are several ways to install the skill:
- Method 1: In the OpenClaw or Hermes Agent chat window, directly tell the Agent: Please help me install the writing-plans skill. (Easiest)
- Method 2: Visit the skillhub website, install the skillhub store first, and then install the corresponding skill. (For Chinese users)
- Method 3: Visit the Skills.sh website, search for the skill name on the homepage, and use the provided command to install it. (For technical users)
- Method 4: Visit the Clawhub website, search for the skill name on the homepage, click the download button to download the zip file, unzip it and place it in the skills directory of OpenClaw.
Skill Workflow Analysis
- Declare Intent and Isolate Environment: Actively announce the use of this skill at startup and enforce running in a dedicated isolated worktree. The generated plan documents are saved to the
docs/superpowers/plans/directory by default. - Mandatory Scope Check: Evaluate whether the current requirement spans multiple independent subsystems. If so, proactively pause and suggest breaking it down into separate plans to maintain modularity at the architectural level.
- Map File Structure and Responsibilities: Before defining specific tasks, outline all file paths that need to be created or modified, and clarify the single responsibility of each file. This step locks in the decoupling strategy early.
- Micro-Task Breakdown (2-5 minute granularity): Compress the execution process into minimal actions. Strictly follow the TDD loop: write a failing test -> verify it fails -> implement minimal code -> pass the test -> commit.
- No Placeholders Rule: Reject any form of “implement later”, “add appropriate error handling”, or instructions lacking specific code. Every step must contain all the actual code and exact terminal commands the engineer needs to execute.
- Standardized Plan Header: Require a fixed Markdown header that clearly defines the goal, architecture approach, and tech stack. It also specifies the sub-agent skills (e.g.,
subagent-driven-development) to be invoked for the subsequent execution phase. - Pre-Execution Self-Review: After writing the plan, cross-check against the original spec for coverage, clean up any remaining placeholders, and verify type and naming consistency across tasks.