[NarryG]

	Files	Added	Deleted	Net
Test	6	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​1113	$\color{#cf222e}{\Huge{\mathbf{−}}}$​12	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​1101
Prod	33	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​635	$\color{#cf222e}{\Huge{\mathbf{−}}}$​218	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​417

ELI5

OMP can ask you a question while it works, but native chat could not display its forms. Supported forms now appear in Orca's existing question cards. You can answer, decline or cancel; Full access never answers these human decisions for you.

What Changed

Opt OMP into form-only ACP elicitation when a handler is available. Translate supported choices, text, numbers and confirmations into existing desktop/mobile cards, then validate answers on the execution host. Keep the question and field instructions visible, including text-only asks. Preserve Other text, exclusive choices, empty values and omitted optional choices. Accept, decline and cancel stay distinct. Unsupported schemas fail explicitly; URL elicitation is not included.

Why

Reusing ACP and the existing question cards avoids another prompt framework. Bounded host validation prevents malformed or conflicting answers from becoming approval. Existing private request/authentication handling remains unchanged.

Linked Issue

Fixes stablyai#26924.

Visual Proof

Windows: before at 51530bb, after at cbee6dff.

Before: same native command, no form cards.

After: OMP multi-select waits for the person under Full access.

Resolved answers, decline and cancel

Testing

• Agent-driven rebuilt Windows app: ten real native form scenarios, including the corrected text-only question; pending form checked in light and dark mode.
• Automated tests added/updated: schema/answer boundaries, empty and omitted values, exclusive/Other choices, decline, cancellation and mobile responses.
• Linux full lint, typecheck and all four pnpm test shards passed at the final head: run.
• Windows full build/typecheck, changed-code quality and targeted question-context/typed-answer regressions passed.
• The broader pre-correction Windows focus had six SQLite fixture-cleanup failures (EBUSY); no fixture workaround or skip is included.
• Native answers were checked with 150 ms added ACP input latency. This is not a real SSH or mixed-peer run.

AI Disclosure

OpenAI Codex GPT-6.1 Sol assisted implementation, integration and review. Earlier independent Standards review used Claude Fable 5.1. Checks were agent-run, not human testing.

Review

Cross-platform: existing Node/TypeScript paths; no new OS-specific behavior. Local/folder/SSH: execution host owns validation and settlement. Agents: OMP opt-in only; other integrations unchanged. Compatibility: optional journal fields and existing answer transport; no new stream opcode or Git-provider assumption. Security/performance: bounded schemas and answers, no automatic approval or credential changes. UI/mobile: existing cards/tokens, visible question context and explicit decline/cancel; no new prompt framework.

Agent skill upstream boundary

• Not applicable; no upstream skill-installer material copied.

Notes

Native timeouts without a withdrawal can leave a pending card; Stop clears it. Rendered checks were Windows-only; real SSH/WSL, mixed-peer, macOS and mobile devices were not tested. No dependency or release-version changes.

Author

GitHub @NarryG · X @NarryOnTheNet.

Checklist

• Focused change, rationale, issue, genuine before/after and review dimensions documented.
• Required contribution checks completed; Windows-specific cleanup failures disclosed above.