| Route | Use when | Load | Notes |
|---|---|---|---|
| Default Harness readiness | The user asks for /better-harness, readiness, a report, recurring friction, or does not name a model. |
agent-work-loop.md directly. |
Use session-rich with reviewed task evidence; otherwise use session-limited and lead with inspected project evidence. Combine the lead, session, and project inputs into five independently scored dimensions without a headline average. |
| Static project scan | The user explicitly requests a repository-only/static score or invokes the quickstart static surface. | software-fluency.md |
Software Fluency owns that explicit static output and may also guide the independent project evidence pass. Session absence alone does not select it for /better-harness. |
| Harness evidence lens | The report needs detailed harness diagnosis, static evidence honesty, agent work-radius scoring, or F0-F4 submetric detail. | harness-engineering.md |
Supporting evidence contract, not a second default maturity score. |
| Repeated Friction Triage | The user reports recurring errors, corrections, skipped review, setup/validation failure, permission or Hook friction, or capability-use gaps. | Harness skills/better-harness/SKILL.md#execution-routing first; then one or more existing owners. |
A trigger router, not a score model. Repetition can trigger Software Fluency inspection, linked Agent Work Loop analysis, Customization Checkup, or owner-gated Skill Discovery; it is not scoring evidence by itself. |
The default model is Agent Work Loop . Session probing sets evidence richness, not model selection:
reviewed Task Episodes use session-rich; partial or absent behavior evidence uses session-limited and
keeps unavailable claims Unobserved. Every durable report launches the session and project passes concurrently.
The project pass may use Software Fluency's five capability questions without creating a second score model. The
lead reconciles both briefs with its neutral evidence brief and authors the retained Agent Work Loop findings
directly. report.source.json is only an evidence envelope, not a scoring owner.
Do not select case-study examples as active built-in models. When the user asks
for Factory-style criteria coverage, migration context, or corpus comparison,
inspect case-studies/factory/model/factory-readiness.md
as reference material and project useful criteria into the selected built-in
model.