doubt-driven-development

SUMMARY

What it does

This Agent Skill, named 'doubt-driven-development', is a process-oriented skill that instructs an agent to subject every non-trivial decision to a fresh-context adversarial review before finalizing it. The skill is designed for use in high-stakes or unfamiliar coding situations where correctness is critical. It defines a five-step process (CLAIM, EXTRACT, DOUBT, RECONCILE, STOP) and includes guidance on when to use or not use the skill, common rationalizations, red flags, and interaction with other skills. The source is a Markdown file (SKILL.md) from the GitHub repository 'addyosmani/agent-skills', fetched and inspected successfully. The skill is provided under the MIT license. No authentication is required, and the skill is free to use. The skill is intended for the main-session orchestrator and explicitly warns against adding it to a persona's skills frontmatter to avoid nested persona invocation. It also includes a mandatory step to offer cross-model review in interactive sessions, with safety precautions for invoking external CLIs. The skill does not define any tools or API endpoints; it is purely a set of instructions for the agent to follow.