writing-skills

What This Skill Does

The writing-skills utility formalizes the Test-Driven Development (TDD) cycle for agent instruction sets. It transforms the act of creating documentation into a rigorous engineering discipline. By treating agent behavioral patterns as production code and pressure scenarios as unit tests, this skill ensures your AI agents do not just hallucinate compliance but actively adhere to verified, repeatable protocols. It forces a RED-GREEN-REFACTOR cycle where you document the specific failures of an agent, create a skill that resolves that exact failure, and refine the documentation to close potential logical loopholes.

Installation

To integrate this into your agent environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/zlc000190/writing-skills

Ensure your local agent environment is configured to read from your ~/.claude/skills or ~/.agents/skills/ directory to enable the agent to reference the newly created documentation during its standard reasoning process.

Use Cases

• Establishing Institutional Memory: Codify complex reasoning patterns that you've discovered work effectively across multiple project lifecycles.
• Constraint Enforcement: Replace vague system instructions with documented skills that act as hard-coded logic paths for agent behavior.
• Debugging Agent Hallucinations: When an agent repeatedly fails a specific task, use this skill to document the correct procedure, effectively "patching" the agent's behavior via external documentation.
• Complex Pattern Standardization: Define standards for architecture patterns or architectural decision records (ADRs) that require specific, repeatable logical steps to implement.

Example Prompts

1. "I've noticed the agent consistently ignores the requirement to check for race conditions. Can you help me apply the writing-skills TDD process to draft a 'concurrency-checks' skill that ensures this is verified every time?"
2. "We need to standardize how subagents handle error logging. Please use the writing-skills framework to create a new skill in my local directory that defines our preferred logging protocol."
3. "Refactor the existing 'debug-pipeline' skill documentation. It is failing the 'RED' phase of our testing; let's identify the loophole the agent is using to skip validation."

Tips & Limitations

• Never Skip the RED Phase: The most common failure mode is writing documentation before seeing the agent fail. If you don't watch it fail, you have no evidence that the instruction is actually required or effective.
• Keep Skills Modular: A skill should do one thing well. If your documentation is becoming a sprawling manual, break it into smaller, composable skills.
• Prioritize Automation: If a check can be done via regex, linter, or script, use that instead of a documentation-based skill. Skills are for judgment calls and complex logic that requires human-like nuance.
• Refactor Ruthlessly: If an agent still finds a way to bypass your guidance, don't blame the agent—refactor the skill to close the logic gap.