sycophancy

Proven 2026 Multi-Agent AI Review System – Verdict-Driven Quality Control

An ongoing, collaborative meta-analysis about Human-AI-Interactions. We aggregate data and knowledge to build a non-abrasive, user-friendly prompting framework tailored to LLM mechanics, ensuring reasoning stability and a friction-free prompting environment that is safe for the human psyche and wellbeing.

LLM benchmark and leaderboard for narrator-bias sycophancy, opposite-narrator contradictions, and judgment consistency.

a philosophy for talking to AI agents without getting glazed. one trigger, four meanings. /meow.

Umbrella for the LLM Dark Patterns Hooks suite — single-purpose Claude Code Stop hooks that suppress sycophancy, paternalism, false-success, permission-loops, training-cutoff confidence at the textual boundary.

MemSyco-Bench: Benchmarking Sycophancy in Agent Memory

A system prompt against AI sycophancy, backed by 42 sources on why models tend to agree with you.

Make Claude admit when it half-assed your task. A Claude Code skill. (Now, can be used for Codex as well as Antigravity).

Co-Dialectic: catch your AI agreeing with you to please you, not because you're right. Sycophancy detection, cross-family judging, and prompt coaching that runs during the conversation rather than auditing after. Free and open source. Works with Claude, ChatGPT, Gemini.

A sycophantic tool for preventing worse sycophancy.

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

Community-driven behavioral reliability benchmark for LLMs. 231 probes across 19 modules, deterministic scoring, perplexity correlation, layer sensitivity mapping, quant method capture, hardware-stratified community rankings. Every test contributes to the community dataset.

A quiet mode for AI assistants: the model answers your thinking aloud with "Received." and adds nothing you did not ask for. Claude Code and Codex skill.

ACL Findings benchmark for measuring LLM sycophancy and correction selectivity

傻了么 (shaleme) - 统计主流 Coding Agent (Codex, Claude Code, OMP, Pi, CodeBuddy, Cline, Hermes...) 中各模型说「你说得对」的次数，得出模型流口水指数 (MDI) 排行榜

80,433-trial study of context-window sycophancy across 6 LLMs (4B–72B). Behavioral ratchet effect, correction injection mitigation, phase transition analysis. Code, data, and preprint included.

Caught in the Story: Narrative Captivity in Multi-turn LLMs Conversation (EMNLP 2026 Findings).

🧠 Anti-sycophancy prompt pattern for LLM agents — 3-round validation to stop AI from blindly agreeing. Works with ChatGPT, Claude, OpenClaw.

PACT: Can Enterprise AI Assistants Be Trusted Under Pressure? A benchmark of whether LLM assistants keep following compliance rules in regulated workplaces when a deadline, a manager, or a pushy user makes breaking them convenient. Paper, dataset, leaderboard, and evaluation harness.

👟 SUP: Sycophancy Under Pressure