meta-optimize

You can optimize prompt workflows by analyzing accumulated usage logs to identify frequent parameter overrides and tool failure patterns, then generating minimal patch diffs for SKILL.md components and convergence rules.

Evidence-based prompt tuning uses real auto-review-loop behavior and tool failure data from execution logs to propose targeted improvements to reviewer gates and workflow defaults instead of manual adjustments.

You need at least 5 logged workflow runs in your events log after enabling Claude Code meta logging to start analyzing tool failures and generating targeted patch proposals for your ARIS project.

No, workflow tuning requires a human-in-the-loop application flow where cross-model adversarial review validates proposed patches first, followed by explicit user approval before applying changes with backups and change logging.

The best way to reduce recurring tool failures is analyzing auto-review-loop behavior and override frequency from real usage logs to rank optimization targets and generate evidence-grounded patches for workflow defaults.

Cross-model adversarial review validates that proposed optimization patches are supported by evidence from usage logs and will not break common workflows, ensuring safer incremental changes before user-approved application.