Автооптимизация рабочего процесса ARIS

★ 7.1 · ai-tooling

meta-optimize is a Claude Code skill that analyzes ARIS usage logs and proposes optimizations for SKILL.md files, reviewer prompts, and workflow defaults — implementing an outer harness improvement loop inspired by Meta-Harness (Lee et al., 2026). It reads execution traces from `.aris/meta/events.jsonl`, identifies bottlenecks across skill invocations and sessions, and stages patch files under `.aris/meta/`. The skill is structurally read-only: it has no Write or Edit tools and cannot apply its own patches — landing changes requires a separate, human-invoked `/meta-apply` skill. Requires at least five logged skill invocations to produce meaningful analysis. Designed for research teams who want to systematically improve their own harness components — prompt instructions, convergence rules, workflow ordering — based on accumulated runtime data.