adaptive-harness

To auto-tune your agent harness, run /omcustom:adaptive-harness to scan the project context, identify required agents, and prune unused configurations automatically. It generates a portable project-profile.yaml that drives future agent activation decisions for the optimized harness.

Project profiling works by scanning the project context to identify required agents and configurations, then merging automated detections with existing profiles to preserve manual edits. It records learned patterns over time and stores results in a portable .claude/project-profile.yaml file for reuse across teams.

Yes, the adaptive harness applies to multiple projects by merging new detections with existing content while preserving manual edits. It writes changes to .claude/project-profile.yaml, ensuring your custom configurations remain intact while applying automated agent activation decisions across different project contexts.

The adaptive harness supports four commands: --scan to detect project context, --optimize to prune unused agents, --learn to record patterns over time, and --export to generate a portable project profile. These commands write changes to .claude/project-profile.yaml while preserving manual edits.

You should avoid automated agent pruning when you have critical manual edits in your project-profile.yaml that rely on agents not detected by the context scan. While the harness preserves manual edits during merging, aggressively pruning unused agents might impact configurations you need for specific edge-case workflows.

The best way to generate a portable project profile is to run /omcustom:adaptive-harness with the --export command after scanning. This produces a .claude/project-profile.yaml file that drives agent activation decisions and can be reused across teams to maintain consistent harness optimization configurations.