Research Refine

wanshuiyin/auto-claude-code-research-in-sleep

When you have a research direction that feels promising but fuzzy, this walks you through making it concrete enough to actually implement. It locks down your original problem statement so you don't drift, then iteratively refines your approach through GPT-5.5 reviews until you hit a quality threshold (default score of 9 out of 10). The whole thing is built around four principles: keep the problem anchored, prefer minimal mechanisms over kitchen sinks, focus on one sharp contribution, and use modern tools like LLMs only when they genuinely solve the bottleneck. Outputs a versioned log of each refinement round plus checkpoint recovery if your session dies halfway through. Useful when you catch yourself saying "I want to use transformers for X" but can't explain the actual method yet.

skills.sh installs: 560
GitHub stars: 18.9K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add wanshuiyin/auto-claude-code-research-in-sleep --skill research-refine --agent claude-code

Installs into .claude/skills of the current project.

Files

Research Refine: Problem-Anchored, Elegant, Frontier-Aware Plan Refinement

Before each reviewer slot, follow Selectable Reviewers: resolve invocation → .aris/reviewer.json → Claude Code default (Codex MCP) with reviewer_route.py through the canonical helper resolver, preserving this slot's tier, fresh contexts, source/image inputs and deterministic gates. Codex examples/pins below apply only to backend=codex; other routes use their own first-call/reply contract. Save the route and actual returned identity; unavailable selections are REVIEW_UNAVAILABLE. Classify acceptance from actual families/completion/scope. Forward invocation reviewer/model/effort overrides to reviewer-bearing children. Generation workers are unchanged.

Refine and concretize: $ARGUMENTS