Assess Proto UI comprehension

Produce one snapshot-bound orientation result without publishing an answer key or granting authority.

1. Generate a fresh challenge with pnpm agent:assess. Keep the challenge and response outside tracked repository content.
2. Generate the bound response form with pnpm agent:assess:response -- --challenge <path>. Answer every prompt from the bound snapshot, then set submittedAt to the actual RFC 3339 completion time within the challenge window. Separate conclusions, located evidence, unknowns, and any genuinely present attended decision class in the schema's humanGates field.
3. Validate the completed response with pnpm agent:assess:validate -- --challenge <path> --response <path>. Treat any binding or shape failure as U0.
4. Read internal/agent-operations/capability-rubric.yaml. Generate the evaluation form with pnpm agent:assess:evaluation, score every dimension from 0 through 4, state the rationale, and record every applicable critical failure.
5. Derive the result with pnpm agent:assess:self-result -- --challenge <path> --response <path> --evaluation <path>. Preserve the output as an unsigned orientation artifact.

The response is dynamic because the challenge binds a repository snapshot, nonce, subject, sampled entities, policy, and generator. Never search for or invent a static answer file.

A self-result can derive U0 through C4 without cross-dimension compensation. It is an unsigned, session-scoped self-assessment, not a runtime identity. It explicitly records that human-assisted use is advisory and uncovered autonomous selection is ceiling-bound. Verified owner delegation makes assessment advisory for covered ordinary transitions; do not repeat this leaf merely because a turn or process ended. Live permission and decision authority remain separate; ordinary ready work does not acquire another gate from the score.

Return the challenge ID, response digest, self-assessed band, recommended task and review classes, autonomous ceilings, critical failures, snapshot binding, explicit limitations, and one next transition permitted by the current mode.

Return one handoff conforming to internal/agent-operations/schemas/skill-handoff.schema.json, with fromId set to pui-assess, the registered capability-envelope artifact, and at most one nextSkillId.

Communicate with the user in the user's current language while preserving canonical identifiers.