derveloper-skills

Use when the user wants to run, monitor, or recover a tmux-pair solo coding workflow through `/run` or `/solo`: one agent in a fresh git worktree, gated recon, plan checks, implementation, final verify, commit hygiene, and auto-squash-merge. Covers durable standards, review-ledger evidence, CLARIFY-NEEDED, plan drift handling, PROJECT.md care, DONE-MERGED, compact recovery, repo-specific subagents, and mandatory post-merge retro.

Adversarial code review mit Dinesh vs Gilfoyle (Silicon Valley). Aufruf: /dg (git diff), /dg 3 (3 Runden), /dg src/foo.rs (Datei), /dg src/foo.rs 3 (Datei + Runden). Zwei Agenten debattieren ueber Code-Qualitaet: Gilfoyle greift an, Dinesh verteidigt. Bestaetigte Issues werden im Final Summary gesammelt.

This skill should be used when the user asks about GEPA, prompt optimization, or text artifact evolution. Typical triggers: "optimize a prompt", "improve a system prompt", "tune a prompt with GEPA", "run GEPA optimization", "evolve a prompt", "optimize anything with LLMs", "create a GEPA evaluator", "optimize MCP tool descriptions", "optimize RAG", "set up GEPA", "auto-tune my prompt". Implements the GEPA algorithm natively in Claude Code without external LLM API keys. Claude Code acts as the reflection and mutation engine.