Und3rf10w (@Und3rf10w) profile picture

Principal Research Scientist @ ReliaQuest by day, security and AI research by night; depending on intoxication levels

Joined July 2012
Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)Photo by Und3rf10w (@Und3rf10w)
This is what peak performance with Opus 4.7 looks like: /plugin marketplace add Und3rf10w/claude-litter /plugin install deepwork@claude-litter >Launch deepwork to <hard goal> This is incredibly expensive, so I would only recommend this for people on an API pricing plan.
2
1
7
376
- No claim without a citation - Cross-check every null with ≥2 agents from different angles - Load-bearing unknowns get tested live - Version every proposal by name, bump on content change - Prior run's incidents become next run's guardrails
1
125
The human steers. Doesn't author or code. I packaged the pattern into a plugin: github.com/Und3rf10w/claude-… Ask it to "spawn deepwork" and give it a goal. CRITIC is the only fixed role. The others get named per problem (auditor, falsifier, scope-cutter, whatever fits the target).
3
116