Model Onboarding Skill¶
The canonical skill is
skills/audio-model-onboarding/.
Codex and Claude resolve to that directory through project-local relative symlinks, so the workflow
and scientific policy remain agent-neutral.
Supported modes are analyze, resolve-environment, integrate, verify, card, and profile.
integrate is a workflow mode, not a lifecycle state. The committed lifecycle states remain
draft, analyzed, environment_resolved, checkpoint_verified, runtime_verified, and
profiled.
Workflow boundaries¶
analyzestatically inspects an upstream project and produces evidence-grounded reports. It does not execute upstream code, download checkpoints, or silently choose a source, variant, checkpoint, or embedding.resolve-environmentderives compatibility candidates from evidence and may run controlled trials only in isolated model-specific environments. Promotion requires successful recreation and verification with exact evidence and failure records.integratemay add a real wrapper, model-specific package code, committed environment and checkpoint specifications, documentation, and tests. It requires an explicitMODE: integraterequest, reviewed analysis, resolved source strategy/variant/checkpoint/environment strategy, a resolved or explicitly deferred primary embedding, and explicit production authorization.verifymay acquire only the explicitly selected checkpoint and execute only the selected model through the existing environment and checkpoint infrastructure.cardcreates or updates one checkpoint-specific card only from validated evidence and completed workflow artifacts.profileis reserved. Profiling remains unavailable until a model isruntime_verifiedand a profiling workflow is explicitly implemented and invoked.
Every mode executes only its own scope. No mode adds model dependencies to the root project, commits checkpoint binaries, silently begins another mode, or creates a Git commit.
Evidence and evaluation¶
The end-to-end workflow is static inspection, structured observations, evidence-grounded analysis, decision gates, compatibility resolution, integration, runtime verification, and model-card authoring. See the evidence policy and artifact guide.
Synthetic evaluation compares each scenario, production-inspector observations, and its golden
analysis report. Reports must cite concrete files, symbols, dependency declarations, revisions,
checkpoint candidates, source strategies, and embedding tensor candidates. Static inspection shares
a bounded InspectionBudget and never executes repository code.
Use the canonical agent request to start a workflow and the canonical response to report results consistently.