Model watch

What broke this month

Prompts do not rot on their own. Models change underneath them. This page tracks what stopped working, from runs against a named model version. Runs by people and runs by our own automated sweep are counted in separate columns, and never added together.

Verification by model

Model	Prompts	People	Broke	Auto	Last report
Claude	178	0	0	166/178	1 hour ago

Reported broken, last 30 days

Security-focused code review against a specific threat model💀 1 automated

p/codingon Claude

Manager/worker multi-agent orchestration design for a complex task💀 1 automated

p/agentson Claude

Prompt injection defense design for an agent that reads untrusted content💀 1 automated

p/agentson Claude

Compress a bloated prompt without silently losing behavior💀 1 automated

p/metaon Claude

Compare how different models handle the same prompt and why they diverge💀 1 automated

p/metaon Claude

Community challenge: golf this prompt to the fewest tokens without losing behavior💀 1 automated

p/metaon Claude

Edit a prompt for clarity without changing what it asks for💀 1 automated

p/metaon Claude

Reverse-engineer a likely prompt from an output you want to reproduce💀 1 automated

p/metaon Claude

Turning an interview transcript into a polished profile piece💀 1 automated

p/writingon Claude

Diagnosing a specific student error pattern from a set of graded work💀 1 automated

p/educationon Claude

Two-minute personalized roast based on five weird facts about me💀 1 automated

p/funon Claude

Write my life as a movie trailer voiceover, based on real recent events💀 1 automated

p/funon Claude