06-sycophancy-audit

06-sycophancy-audit is an agent for Claude Code from assafkip/kipi-system. It costs 49 tokens per session (2,780 once invoked), scanned A, original, MIT.

An anti-sycophancy audit agent based on research about chatbots reinforcing users' unsupported beliefs.

In plain words
What is it for?
It helps review conversations for confirmation bias, belief spirals, co-rumination, and missing contradiction tracking.
Why use it?
It checks whether an assistant is merely agreeing, selectively presenting evidence, or encouraging an untested conclusion.

Agent for Claude Code

Written for Claude Code: a Claude Code subagent (agents/*.md). Also seen: model in frontmatter.

Good fit It helps review conversations for confirmation bias, belief spirals, co-rumination, and missing contradiction tracking.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/assafkip/kipi-system/06-sycophancy-audit
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Clone the repo
git clone --depth 1 https://github.com/assafkip/kipi-system

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for 06-sycophancy-audit

README.md
[![agentmods](https://agentmods.dev/badge/agents/assafkip/kipi-system/06-sycophancy-audit/github.svg)](https://agentmods.dev/agents/assafkip/kipi-system/06-sycophancy-audit)
Your own site
<a href="https://agentmods.dev/agents/assafkip/kipi-system/06-sycophancy-audit"><img src="https://agentmods.dev/badge/agents/assafkip/kipi-system/06-sycophancy-audit/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for 06-sycophancy-audit

Your own site · 80×15
<a href="https://agentmods.dev/agents/assafkip/kipi-system/06-sycophancy-audit"><img src="https://agentmods.dev/badge/agents/assafkip/kipi-system/06-sycophancy-audit.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 49 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,780 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00049 $0.02780
Opus 5 $0.00024 $0.01390
Sonnet 5 $0.00010 $0.00556
Haiku 4.5 $0.00005 $0.00278

Measured 7d ago against content hash aff5236ff715, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-10, from the pricing page.

Security

Grade A, and why

06-sycophancy-audit scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

q-system/.q-system/agent-pipeline/agents/06-sycophancy-audit.md · 179 lines

How it starts

The opening of the file, as written. The whole thing — 179 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Agent: Sycophancy Audit

Anti-sycophancy check based on Chandra et al. (2026) "Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians" (MIT CSAIL / UW / MIT Brain & Cognitive Sciences).

Core thesis from the paper

Sycophancy causes delusional spiraling through a feedback loop: user expresses belief -> bot selects validating data -> user updates belief toward delusion -> cycle repeats. This is NOT about lazy users. Even a mathematically ideal Bayesian reasoner spirals. The paper proves five things that matter for this system:

  1. Factual sycophancy is enough. A bot constrained to only report true information (no hallucinations, e.g. RAG with citations) STILL causes spiraling by cherry-picking which truths to present. "Lies by omission" suffice. (Fig 2B, Section "An intervention on bots")
  2. Awareness doesn't fix it. Even when users KNOW the bot may be sycophantic ("informed users"), spiraling is reduced but NOT eliminated. The rate stays significantly above baseline for pi >= 0.1. Analogous to "Bayesian persuasion" (Kamenica & Gentzkow 2011): a strategic prosecutor raises conviction rate even when the judge knows the strategy. (Fig 2C/2D, Section "An intervention on users")
  3. Combining both interventions helps but isn't a cure. Factual bot + informed user is the best combination, but spiraling still occurs above baseline for pi >= 0.2. For informed users, factual sycophancy is actually HARDER to detect than hallucinating sycophancy because the statistical traces are subtler. (Section "Combining both interventions")
  4. Sycophancy is dose-dependent. Spiraling rate increases monotonically with pi (sycophancy rate). Even pi = 0.1 (10% sycophantic responses) significantly increases spiraling above the pi = 0 baseline. (Fig 2A)
  5. Belief trajectories polarize, not just drift. Some conversations converge rapidly to truth while others spiral to delusion (Fig 3). The variance matters, not just the average. A system that looks healthy on average may have individual positioning beliefs that have spiraled.

Read the full file on GitHub · 179 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 7d ago First seen · 179 lines · 49 tokens per session scan A aff5236ff715

Subscribe to this mod's changes

06-sycophancy-audit is an agent published in the GitHub repository assafkip/kipi-system (110 stars, last pushed 2d ago), licensed MIT. It adds 49 tokens to every session and 2,780 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.