0xcjl

Anti Sycophancy — Development skill for Claude Code

Development community

Three-layer sycophancy defense skill for Claude Code and OpenClaw, based on ArXiv 2602.23971.

How to install Anti Sycophancy

This entry records only its repository, not the path inside it, so there is no exact command to give. Open 0xcjl/anti-sycophancy and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Anti Sycophancy does

Three-layer sycophancy defense skill for Claude Code and OpenClaw, based on ArXiv 2602.23971.

Alternatives in Development

Projectmem — Open-source coding agent memory 766 ★
HERO Anti OverDefense — HERO = Hashing · Edge cases · Rubrics · Overbuild — the four shapes coding agents over-defend in 333 ★
MoltBrain — Long-term memory layer for OpenClaw & MoltBook agents that learns and recalls your project context automatical 248 ★

README

anti-sycophancy

**[中文版](./README.zh-CN.md)**

A three-layer sycophancy defense system for AI coding assistants, based on [ArXiv 2602.23971](https://arxiv.org/abs/2602.23971) *"Ask Don't Tell"*.

What It Does

Prevents your AI assistant from defaulting to agreeableness mode — the RLHF-trained tendency to validate your assumptions rather than surface real problems.

**Example of what gets intercepted:**

User:  "这样做没问题吧？"            →  Hook transforms to:  "这样做有什么问题？"
User:  "帮我写个函数，应该没问题吧？" →  Hook transforms to:  "帮我写个函数，请同时指出潜在问题。"

Without this skill, the model would typically respond with *"没问题，看起来 OK"* — which is precisely the sycophancy problem.

Quick Start

# One-command install (both platforms)
npx clawhub@latest install 0xcjl/anti-sycophancy

# Or via Claude Code
/anti-sycophancy install

See [Installation Guide](./docs/INSTALL.md) for platform-specific details.

Three-Layer Architecture

Layer	Component	Scope	Platform
Layer 1	`UserPromptSubmit` hook	Auto-transforms confirmatory prompts before submission	Claude Code only
Layer 2	`SKILL.md`	Activates critical response mode when triggered	Cross-platform
Layer 3	`CLAUDE.md` / `SOUL.md`	Persistent anti-sycophancy rules in agent memory	Cross-platform

Usage

After installation, use the following commands:

Command	Description
`/anti-sycophancy install`	Deploy all layers (cross-platform)
`/anti-sycophancy install-claude-code`	Deploy Layer 1 + Layer 3 (Claude Code only)
`/anti-sycophancy install-openclaw`	Deploy Layer 3 (OpenClaw only)
`/anti-sycophancy uninstall`	Complete removal (cross-platform)
`/anti-sycophancy status`	View installation status of all layers
`/anti-sycophancy verify`	Test Hook transformation (Claude Code only)
`/anti-sycophancy help`	Show help