Convert a raw specification into a polished implementation plan through iterative review passes.
npx skills add https://github.com/dreambound/plan-refiner-skill --skill plan-refiner
Please help me install this Agent Skill. Skill: plan-refiner Source: https://github.com/dreambound/plan-refiner-skill/tree/main Command: npx skills add https://github.com/dreambound/plan-refiner-skill --skill plan-refiner
This Skill transforms a starting specification into a polished implementation plan by orchestrating iterative refinement passes with a fresh agent perspective, reducing drift and misalignment across iterations.
Provide an initial specification or prompt to begin refinement. The agent will slug the spec, create a namespaced workspace under ~/.claude/plans/plan-refiner, generate an initial plan, and guide you through iterative passes with user-in-the-loop finalization.
The skill implements an iterative refinement process with a generation step, a review loop (minimum 3 passes), user prompts for clarifications, and a finalization decision. Reviewers use Context7 to verify current versions, and a changelog documents each pass.
High-intent search queries and answers about installing and using this skill.
You can refine an implementation plan by orchestrating iterative passes with fresh agent perspectives, re-reading the original spec each time to prevent drift. This process standardizes plan refinement by generating an initial draft and running parallel standard, adversarial, and custom reviews.
Iterative plan refinement is a process that generates an initial plan from a specification and then improves it through multiple review passes. Using fresh agent perspectives for each pass surfaces diverse feedback and reduces misalignment across iterations until the user finalizes the plan.
You can run parallel adversarial and standard reviews by providing an initial specification to the refinement process. The system orchestrates these parallel review passes simultaneously to accelerate feedback convergence and surface diverse issues before finalizing the plan.
Yes, the plan refinement process uses Context7 to verify current library versions and ensure version compatibility during each review pass. This verification happens automatically as reviewers evaluate the generated implementation plan.
Yes, the refinement process preserves a full audit trail at ~/.claude/plans/plan-refiner/{spec-slug}/. This audit directory maintains all plan versions, pass changelogs, and reviewer feedback to support complete traceability of the iterative process.
You need to capture user clarifications when the review process surfaces issues or ambiguities in the current plan. The user-in-the-loop approach prompts for clarifications during each pass and lets the user decide exactly when to finalize the polished implementation plan.
sadie100Draft and iteratively refine PRDs and design documents through collaborative questions.
masta-g3Execute active plans and update tracked feature statuses in features.yaml.
CityBear3Coordinate parallel adversarial agents to review code changes and output a unified triage report.
SAZO-KRLaunch five parallel fresh-context code review agents until all pass.
