By WeZZard
Capture and apply human-ruled knowledge the model cannot internalize — user preferences and output-style rulings, settled by oracle: research the world, ask the human, experiment on the unknown.
Settle a question that only trying can answer — an output-style comparison, a prompt phrasing, an error message, any unknown with comparable candidates — by a blind comparison. Produce candidates, strip provenance, collect blind verdicts against one stated criterion, then present everything to the user for the ruling. Use when the attune explore skill classifies an unknown as NOBODY-owned, or when the user asks to settle a question by evidence.
<EXTREMELY_IMPORTANT>You MUST use explore when discussing solutions, designs, or approaches with the user, and whenever a user preference or output-style judgment surfaces. It routes every open unknown to its oracle — research the world, ask the human, experiment on the unknown — so discussions settle on evidence and rulings instead of assumptions.</EXTREMELY_IMPORTANT>
The standard for calling work done — a result is done only when it has survived its use path, driven the way a human would use it, through its real interface. Use before declaring any nontrivial change done, when reporting results, or when deciding how much verification a change deserves. Complements the verify skill where one exists; this skill sets the standard, that one carries the procedure.
Attune injects your standing guidelines into every session — how to write
to you and what counts as verified work — and ships the oracle method for
settling unknowns by evidence. One repo ships three plugins: Claude Code,
Codex, and Pi, all fed by the same references/*.md source of truth.
Claude Code always carries the full feature set; porting.json (the port
matrix) decides what ships to the other platforms. External-agent
delegation lives in the sibling
dispatch plugin — attune's
experiment and verification widen through it when it is installed, and
degrade gracefully without it.
/plugin marketplace add wezzard/skills
/plugin install attune@wezzard-skills
Start a session. Two SessionStart hooks inject the communication and
writing-style guidelines; the skills are attune:explore,
attune:experiment, and attune:verification. Install
dispatch@wezzard-skills alongside for the external-agent router, the
audit panel, and image generation.
codex plugin marketplace add WeZZard/skills
codex plugin add attune@wezzard-skills
Then one one-time step: hooks need your review before they run. Start
codex, and in the "Hooks need review" dialog choose Trust all and
continue (or run /hooks and press t). From the next session on, the
ported hooks inject the communication and writing-style guidelines. The
ported skills are attune:explore, attune:experiment, and
attune:verification (in variants that run without external agents — see
/skills); the external-agent router stays Claude-only per
porting.json.
pi install https://github.com/WeZZard/attune
Start a session. The attune extension appends the ported guidelines
(communication, writing style) to the system prompt at session start. The
ported skills are explore, experiment, verification (in variants
that run without external agents), and keystone; the external-agent
router stays Claude-only per porting.json.
MIT — see LICENSE.
Based on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub wezzard/skills --plugin attuneWeZZard's Claude Code skills collection
Tools for auditing and improving Claude Code skill definitions
Delegate tasks to external agent CLIs through one router and one brief contract — probed facts before use, write isolation through worktrees, a multi-model audit panel, and concurrent image generation with failure memory.
Distilled by WeZZard in China, Assembled from WWDC.
Ultra-compressed communication mode. Cuts 65% of output tokens (measured) while keeping full technical accuracy by speaking like a caveman.
Memory compression system for Claude Code - persist context across sessions
Multi-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
45% cost reduction measured. Cache expiry prevention, SubTask auto-delegation, zero-cost context restoration, real-time cost dashboard. The only Claude Code plugin built from CC source analysis.