fix review findings: polish ask-user question

This commit is contained in:
Yichen Jiang
2026-07-07 14:57:06 +08:00
parent 510e80e447
commit cebf781d69
51 changed files with 665 additions and 419 deletions

View File

@@ -10,19 +10,19 @@ This is a user-facing capability, but it also crosses package boundaries. A mode
## Decision
Introduce `dsh-user-interaction` as the core interface package for `ctx.userInteraction`, and keep the model-facing consumer `dsh-tool-ask-user` under `packages/ui/tool-ask-user` rather than the core spine. The split is intentional: core owns the abstract seam and stable request/answer/error vocabulary; UI product surfaces own the affordance that asks a human and the concrete provider that collects the answer. The tool registers `ask_user_question`, forwards `{ question, header, options, allowCustom, agent, signal }`, and returns the provider-computed `answer` as the tool result.
Introduce `dsh-user-interaction` as the core interface package for `ctx.userInteraction`, and keep the model-facing consumer `dsh-tool-ask-user` under `packages/ui/tool-ask-user` rather than the core spine. The split is intentional: core owns the abstract seam and stable request/answer/error vocabulary; UI product surfaces own the affordance that asks a human and the concrete provider that collects the answer. The tool registers `ask_user_question`, forwards `{ questions, agent, signal }`, and returns the provider-computed structured answers as the tool result.
The request vocabulary supports a short `header`, the required `question`, optional mutually exclusive `options`, `description` for each option, a `recommended` marker, and `allowCustom`. `label` is user-facing display text; `value` is the model-facing answer for a selected option and defaults to `label`. Providers return `AskUserQuestionAnswer.answer` as the single source of truth; the selected `option` is metadata. The tool schema exposes `description` only, not the synonym `desc`, to keep the model-facing surface small.
The model-facing request vocabulary is deliberately aligned with the product-research schema: `ask_user_question({ questions: [{ id, question, header?, options?: [{ label, description? }], multi_select? }] })`. `id` is supplied per question and echoed in the result so a batch can be routed without relying on question text. `label` is both user-facing display text and the selected value returned to the model; there is no separate `value`, no `recommended`, no `allow_custom`, and no `desc` alias.
Optionless questions are always free-form, even if a caller passes `allowCustom: false`. The opposite would create an unanswerable prompt: with no option to select and free-form input disallowed, every human answer would be rejected forever. Providers therefore treat "no options" as the free-form shape.
Providers return `{ answers: [{ id, selected, custom? }] }`. `selected` is always an array of selected option labels, so single-select and `multi_select` answers share one result shape. `custom` carries a free-text "Other" answer; optionless questions collect `custom` directly. When `custom` is present, it overrides any selected choices and `selected` is empty.
`UserInteractionError` extends `HarnessError`, so failures such as `NO_PROVIDER`, `ASK_ABORTED`, ACP cancellation, or missing session routing survive `ctx.tools.execute()` as machine-routable `{ name, code }` tool errors. This matches the structured-error taxonomy and lets the model or a wrapping plugin distinguish "user cancelled" from a generic thrown exception.
## UI mappings
`dsh-stdio-agent`'s in-package readline module renders the question, sorts recommended options first, shows each option's `description` on the next line, accepts the recommended option on an empty answer, and rejects pending questions on abort, provider disposal, or stdin EOF. The stdio provider serializes multiple simultaneous questions with an internal queue so only one prompt owns stdin at a time.
`dsh-stdio-agent`'s in-package readline module renders each question, shows each option's `description` on the next line, supports comma/space-separated numeric choices for `multi_select`, accepts free-form custom answers, and rejects pending questions on abort, provider disposal, or stdin EOF. A batched request is asked in order and resolved as one answer object. The stdio provider serializes simultaneous requests with an internal queue so only one prompt owns stdin at a time.
`dsh-acp` provides the same seam for ACP sessions. It routes an ask request from the calling `Agent` through the bridge's `agent→sessionId` reverse map and calls ACP `unstable_createElicitation` with a session-scoped form. Option choices become a `choice` single-select field with the recommended option as the schema default; free-form answers use `answer` for optionless questions and `custom_answer` when options plus custom input are allowed. ACP `decline`/`cancel`, a missing answer, a missing session, and a client without elicitation support all become structured `UserInteractionError`s.
`dsh-acp` provides the same seam for ACP sessions. It routes an ask request from the calling `Agent` through the bridge's `agent→sessionId` reverse map and calls ACP `unstable_createElicitation` with a session-scoped form for each question. Single-select options become a `choice` string enum; `multi_select` options become a `choice` array enum; optionless questions use a required `custom` text field. If the client returns both `choice` and non-empty `custom`, the custom answer wins. ACP `decline`/`cancel`, a missing answer, a missing session, and a client without elicitation support all become structured `UserInteractionError`s.
The ACP mapping deliberately uses elicitation, not `session/request_permission`. `request_permission` is still reserved for the separate permission gate: it is a yes/no-or-policy authorization protocol around tool execution. `ask_user_question` is a general information-gathering tool with optional free-form answers, so ACP form elicitation is the closer protocol fit. The bridge's session routing is shared with the future permission gate, but the user intent is different.
@@ -46,4 +46,4 @@ The feature gives the model a powerful pause primitive, so prompt guidance matte
## Testing
Unit coverage pins provider registration/disposal, duplicate-provider rejection, abort-before-provider, structured tool errors through `ctx.tools.execute()`, option labels/values, and the model schema including the removal of `desc`. `dsh-stdio-agent` tests cover recommended-first display, descriptions, queued questions, EOF/abort cleanup, and optionless free-form input even with `allowCustom: false`. ACP bridge tests drive a real in-memory ACP connection with the real `ask_user_question` tool and verify both selected-option and optionless free-form elicitation paths continue the agent loop.
Unit coverage pins provider registration/disposal, duplicate-provider rejection, abort-before-provider, empty-question rejection, structured tool errors through `ctx.tools.execute()`, batched answers, multi-select answers, custom answers, and the model schema including the removal of `value`, `recommended`, `allow_custom`, and `desc`. `dsh-stdio-agent` tests cover option descriptions, queued requests, EOF/abort cleanup, optionless free-form input, invalid option reprompts, duplicate multi-select numbers, and batched question flows. ACP bridge tests drive a real in-memory ACP connection with the real `ask_user_question` tool and verify selected-option, custom-overrides-choice, multi-select, and optionless free-form elicitation paths continue the agent loop.