2026-07-26 - Claude 5 Context and Loop Engineering
2026-07-26 - Claude 5 Context and Loop Engineering
Added
- trq212-claude-5-context-engineering - Thariq (@trq212) shares Anthropic's finding that cutting ~80% of Claude Code's system prompt for Claude 5 models caused no measurable eval loss — old "rules/examples/upfront" context engineering has become anti-patterns in favor of judgement, interface design, and progressive disclosure (
claude doctorrightsizes skills/CLAUDE.md). - loop-engineering - Addy Osmani's essay arguing you should design loops that prompt your agents (automations, worktrees, skills, connectors, sub-agents + memory) rather than prompt them turn-by-turn — but verification, comprehension debt, and "cognitive surrender" stay on you.
- brittany-ellich-loop-engineering - Brittany Ellich's real-world case study: an agent loop drove 108 PRs in ~8 days (vs. 5-10/week) via a protocol/loop/worker design with a markdown task-board, self-stopping
/loop, and capped memory. Builds on loop-engineering. - moc-loop-engineering - New Map of Content gathering loop-engineering practices, essays, and people (loop-engineering, brittany-ellich-loop-engineering, and adjacent skills/agents/context-engineering entries).