Google WikiSkill: Persistent Wiki Improves Skill-Evolving Agents
rohanpaul_ai · x · 2026-08-29
Google presents WikiSkill, an architecture that improves skill-evolving agents by maintaining a persistent wiki memory.
The Problem: Skill-evolving agents rewrite skill files every round, scattering the reasoning behind edits and losing rejected proposals.
The Solution: WikiSkill splits the agent workspace into three layers:
- Immutable execution traces
- Active skill file
- A Wiki layer (in between)
The wiki records failure patterns and the outcome (accept/reject) of every past proposal. If a skill edit is rejected, it disappears from the active file but remains in the wiki.
Results: On Gemini-3.5-Flash, WikiSkill averaged 68.1% across 5 benchmarks, beating the strongest competing method (56.1%) and the no-skill baseline (49.5%).
Caveat: Skills are handed to the agent rather than retrieved, so skill selection is not part of the evaluation.
More from coding & agent
- How to hook Claude Code or Codex up to iMessage in 3 steps — jacob_posel · 2026-10-16
- Are the fastest GPU kernels now written by AI? Kernels are hyper-verifiable, argues one dev — fleetwood___ · 2026-10-16
- Anthropic lead engineer's 4-question prompt makes agents more ambitious, now a free skill — daniel_mac8 · 2026-10-16
- ChainForge dev shows 'feature map' for orchestrating subagents on large PRs — IanArawjo · 2026-10-16
- aiDotEngineer benchmarks agent orchestration pipelines, staging humans as Foreman and workers — vikvang1 · 2026-10-16
- Developer mass-produces 200 videos in hours using Claude Code + Hyperframes — mcraddock · 2026-10-16