Plugin

llm-dark-patterns

A 10-hook suite that suppresses LLM dark-pattern defaults at the textual boundary. Bundles three branches: interaction-style hooks (no-vibes false-success closeouts, time-anchor training-cutoff confusion, no-curfew unsolicited paternalism, no-sycophancy praise-spam at turn open, no-cliffhanger dangling permission-loop endings, honest-eta vibe time estimates plus linear-scaling claims), fact-fabrication hooks (no-fake-recall false-memory claims without quoted prior content, no-fake-stats fabricated percentages without source, no-fake-cite academic citations without verifiable URL), and one continuity hook (no-amnesia, snapshots working state on Stop / PreCompact / PostCompact, rehydrates on SessionStart so context survives auto-compaction). Each hook is conservative on purpose with explicit allow-clauses documented in its RECEIPTS.md. Backed by AbstentionBench, Anthropic's Constitution, AAAI 2026 sycophancy research, the CoT-faithfulness literature, and the Frontiers in AI 2026 paper on LLM-mediated cost drivers. Includes a METHODOLOGY.md harness-engineering playbook for shipping new hooks against newly discovered dark patterns.

• Published source
• Repository

Analysed

11 October 2026

Last observed

11 October 2026

What the code does

• Runs shell commands Code that starts system commands on the machine it runs on.
• Runs hooks automatically Hooks that execute without being called explicitly.

Having a capability is not wrongdoing; many tools need these to do their job. It tells you what to check before you install.

This is public metadata, not a safety certification or complete assessment. Static analysis cannot establish what a tool does at runtime.