skill-tuning

To fix context explosion and long-tail forgetting in multi-phase agent skills, you need to diagnose anti-patterns in context handling and memory retention. Universal workflow diagnosis detects these issues and applies targeted optimizations with rollback capabilities to restore state consistency.

The best way to diagnose broken state consistency and fragile agent coordination is through pattern-based workflow checks that detect anti-patterns across memory, constraints, and agent call robustness. Deep root-cause analysis triggers when standard checks are insufficient.

To optimize token consumption and constraint retention for autonomous workflows, run a diagnostic workflow that analyzes token growth and state management. It produces prioritized fix proposals and verifies improvements through quality-gate validation.

Yes, skill tuning works with YAML-defined entry metadata and state-driven orchestration. It requires these structures to perform universal diagnosis, apply fix strategies, and execute quality-gate verification with safe backup and rollback semantics.

Yes, you can use Gemini CLI for deep architecture and performance reviews when pattern checks fail. Gemini-assisted deep analysis triggers on critical or uncovered areas to perform root-cause reviews when standard diagnostics are insufficient.

Your multi-phase skill fails output validation and documentation drift checks due to anti-patterns in state management and tool usage. Diagnostic categories identify these failure modes, enabling you to apply fixes and verify reliability through re-diagnosis.