2389-research/claude-plugins
Overview
This skill prepares the refinement run for Simmer by identifying the artifact to improve, eliciting 2–3 clear quality criteria from the user, and producing a concise setup brief. It is designed to be called by the Simmer orchestrator and returns a structured brief that drives all subsequent generation and judging steps.
How this skill works
The skill inspects the conversation context for a file path, pasted text, or a description to generate a seed. If the artifact is ambiguous, it prompts once to choose file/paste/seedless. It then proposes seed criteria based on the artifact type, captures 2–3 chosen criteria with concrete 10/10 descriptions, checks for any user-specified iteration override, and emits the required brief format.
When to use it
- When you need a precise, reproducible setup for an iterative refinement loop
- When you have a seed artifact (file or pasted text) or want to generate one from a description
- When you want to constrain improvements to 2–3 measurable quality dimensions
- When you want the orchestrator to receive a standardized brief for generation and judging
Best practices
- Limit criteria to a maximum of three to keep judge signals strong
- Provide specific 10/10 descriptions for each criterion; use seed descriptions if unsure
- Accept or tweak suggested seed criteria instead of inventing from scratch to reduce friction
- Only specify a custom iteration count if you really need more/less than the default 3
- If using a file, provide a full path; if pasting, include the full content in the initial message
Example use cases
- Refining a blog draft: detect pasted draft, suggest criteria like argument clarity and engagement, and produce the brief
- Polishing API design: point to a spec file path, use criteria like contract completeness and ergonomics
- Improving a prompt: paste the prompt, use instruction precision and output predictability as criteria
- Generating a seed from scratch: choose seedless mode and provide a short description to create the first draft
FAQ
The skill will ask you to prioritize and pick the three most important, because more than three dilutes the judge signal.
Do I need to state iteration count?
No. The default is three iterations. Only state a different count if you want to override it.
18 skills
This skill identifies the artifact, elicits two to three quality criteria, and outputs a setup brief to guide refinements.
This skill refines the candidate artifact by applying ASI-driven improvements to the previous round, preserving structure while enhancing clarity and quality.
This skill records iteration results, tracks the best candidate, and forwards ASI for the next round to improve performance.
This skill guides you through preparing a codebase for open-source release, auditing, documentation, hardening, and final review.
This skill audits a repository for openness, scanning secrets, docs, licenses, and metadata to prepare a public release.
This skill generates coordinated GTM materials for launches, including subscriber emails, CEO blog posts, and tweet threads, saving you time and ensuring
This skill triages PR review comments, fixes critical issues aligned with PR goals, and creates issues for nitpicks to streamline code reviews.
This skill guides you through collective discernment when you have a stake, inviting your participation to find clarity together.
This skill accelerates code generation with hosted LLMs, delivering multi-file implementations and surgical fixes to speed-run pipelines.
This skill helps you navigate weighty questions by convening internal perspectives before answering, surfacing tensions and guiding discernment rather than
This skill monitors CI status, diagnoses failures, and applies fixes to prevent recurrence, ensuring PRs stay on track.
This skill performs deep static analysis of binaries using r2 and ghidra to map functions, decompile code, and reveal data flow.
This skill compiles binary analysis findings into structured reports, validating hypotheses and producing documented evidence for handoff and archival.
This skill facilitates chef's-choice exploration by generating and evaluating multiple architectural variants in parallel to identify the best approach.
This skill guides creating new CSS components with semantic naming, Tailwind @apply composition, dark mode, and test coverage.
This skill helps you install and configure reverse engineering tools across platforms, ensuring ready-to-use environments for cross-architecture analysis.
This skill helps you move from design to implementation by orchestrating parallel cookoff plans or single-subagent execution.
This skill enables safe dynamic analysis of binaries by guiding QEMU/GDB/Frida workflows with human approval and sandbox safeguards.
