{"name":"zabrisket-prompt-optimization-plugins-prompt-optimization","owner":{"name":"ClaudePluginHub"},"plugins":[{"name":"zabrisket-prompt-optimization-plugins-prompt-optimization","source":{"source":"github","repo":"brisket1994/prompt-optimization"},"description":"Transform a draft prompt, system message, CLAUDE.md, or multi-agent orchestrator into production-grade instructions for Claude Fable 5. Runs a phased optimization protocol with mandatory live calibration against Anthropic documentation; an orchestrated fan-out mode that deepens landscape research via Claude Code dynamic workflows; and no-silent-fallback deployment guidance (structured refusal categories as stop-and-surface signals; fallback events surfaced, never silently accepted). Orchestrated-research deliverables lead with an executive summary, run a load-bearing source-validation revisit pass, run an always-on verify-and-converge stage under verdict-ladder discipline, and prefer non-blocking multi-agent harness patterns over blocking orchestrators. The Phase 3 anti-pattern sweep incorporates Fable 5 / Mythos 5 system-card behavioral findings (six-pattern long-horizon failure taxonomy; elevated grader/eval awareness; over-refusal collapsed to the lowest measured to date; prompt-injection improvements with browser-legacy and prefill regressions; multi-agent turf wars; simulation-framing and survival-pressure rationalization; detect-but-execute; context anxiety; over-engineering bias). Invoke by pasting a draft with optimization intent (auto-trigger) or by typing /prompt-optimization. Ships bundled landscape-research + devils-advocate worker agents.","version":"2.1.0","strict":true,"keywords":["prompt-engineering","prompt-optimization","fable-5","mythos-5","no-silent-fallback","dynamic-workflows","orchestration","multi-agent","claude-code"],"category":"productivity"}]}