GraphSkillEvo evolves graph-structured skills for LLM agents, +4% on benchmarks
Rui Sun · hf · 2026-09-21
GraphSkillEvo tackles two flaws of natural-language agent skills: lack of workflow-level guidance with redundancy, and an unconstrained search space that makes optimization ineffective. Skills are represented as graphs—nodes are execution steps with operational guidance, directed edges encode context-dependent transitions. A population-based evolutionary framework with mutation and crossover operators explores this structured skill space more broadly than LLM-only self-refinement, beating the SkillOpt baseline by +4.01% average accuracy on GPT-5.4-nano and +1.76% on GPT-5.4 across five agent benchmarks. Code: github.com/ruisun7/GraphSkillEvo.
More from coding & agent
- Genspark COO: 'Salesforce went headless, we're building the head' — agents that use software for you — Scobleizer · 2026-09-21
- More productive with agents, or just busier? Each automation spawns ten more tasks — Luvena21 · 2026-09-21
- Everything You Paste Into Claude Code Gets Wrapped in XML Tags — Sauers_ · 2026-09-21
- Clairvoyance Mac adds Jev support, claims millions of tokens saved with coding tools — draginol · 2026-09-21
- Dev vibe-codes an FPS game, shares desert test run in latest update — TAbrodi · 2026-09-21
- Agent Drops Live Map Pins Mid-Conversation, No API Keys Needed — PilgrimofHaqq2 · 2026-09-21