Wishing Willow — AI skill for Claude Code
Show what the model thinks you asked, next to what you actually said.
How to install Wishing Willow
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open yha9806/wishing-willow and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Wishing Willow does
Show what the model thinks you asked, next to what you actually said. A Claude Code plugin that surfaces task-decoding drift — no scoring, no blocking, records stay on your Mac.
Alternatives in AI
- Iai Personal Memory Engine — A cyber brain for your AI 866 ★
- Visual Skills — AI film director skills for agents: cinematic dramaturgy (Murch, blocking, montage) + exact prompt syntax for 212 ★
- AI Surface — Find and govern AI attack surfaces in application code, at PR time and inside your AI coding tool (MCP server 131 ★
README
Wishing-Willow
Shows what Claude thinks you asked, next to what you actually said — and keeps a list of the whole conversation.
A Claude Code plugin, a small macOS app, and lintel, which draws both on the notch.
English · 简体中文
The moment Claude writes how it read you, the notch shows your words and its reading together. You asked for exponential backoff; it read "rewrite the whole retry logic, and the timeout too". It did not mark this as a mismatch. Seeing the gap is up to you. Every session on this page is made up.
I built Wishing-Willow because Claude sometimes reads my request a little differently from what I meant, and the reply reads well enough that I don't catch it for several turns. At the start of each turn a hook saves my prompt verbatim on my machine and asks Claude to write one line on how it read it. When that line appears, the notch shows my words and Claude's reading side by side.
A longer conversation has a second problem: what is still open. Claude keeps a list for the whole conversation — what it is doing, what waits on me, what waits on something else — and changes it only with lines I can see. The notch keeps the count of things waiting on me.
It doesn't score anything, it doesn't block anything, and its records stay on your Mac. Whether the reading matches is yours to judge. That is the whole design.
Why I built it
In the conversation that produced this plugin, the same thing happened three times. The first time, I asked for "the general logic behind this class of problem, something reusable"; Claude read it as "audit this specific repository for evidence
Related Skills
RayCodes AIJobSearch
AI Job Search Github: This Local AI Agent Applies to Jobs For You! (Claude Code / Antigravity) - Local-first a
Agent Eval Hub
Multi-provider + cross-device reliability harness for AI agents. Runs the same suite across Claude, OpenAI, Ge
Willow
Willow — a companion narrator that gives blind developers situational awareness while an AI coding agent (e.g.
Local LLM Benchmarks
Measured llama.cpp benchmarks on AMD Radeon RDNA4 with ROCm: RX 9070 XT + Radeon AI PRO R9700 (48 GB). Qwen3.8
Yunshu
Fast local LLM / VLM inference engine for Apple Silicon (MLX). OpenAI- and Anthropic-compatible, lossless spec
Crab Code
🦀 Open-source alternative to Claude Code, built from scratch in Rust. Agentic coding CLI — thinks, plans, and
Related Agents
Bonsai
Runs a task on the local Bonsai 2 27B model (PrismML ternary 27B, MLX, on this Mac) and returns its answer. Us
Meeting Qualifier
Use when someone shows interest, asks about price or how it works, or asks for a call. Decides book, ask or pa
Verifier Opus
Independent verification of a completed change — does the diff actually do what the plan said, do build/tests/