@cartanova/qgrid-cli 2.9.10 → 2.10.1

package/skills/qgrid/references/ai-sdk-provider-contract.md

@@ -68,7 +68,7 @@ Do not write `providerOptions: { ... } satisfies QgridProviderOptions`: the expo
 - `reasoningSummary`: OpenAI/Codex route only.
 - `serviceTier`: OpenAI/Codex route only.
 - `timeoutMs`: Anthropic server-side Claude Code process timeout in milliseconds. It must be a positive integer no greater than 30 minutes and defaults to 240 seconds. For non-stream `generateText`, the SDK derives request-scoped Undici `headersTimeout` and `bodyTimeout` values as `timeoutMs + 60_000` without changing the process-global dispatcher.
-- `fallbackModels`: reserved for future qgrid server-side fallback routing. It is not the Fable 5 safety-refusal fallback, which is owned by Claude Code upstream.
+- `fallbackModels`: deprecated, ignored compatibility field. The provider emits a standard unsupported-option warning when supplied; remove it from callers. It is not Claude Code's upstream Fable safety-refusal fallback.
 - `imageGeneration`: OpenAI/Codex non-stream only. Enables Codex's built-in `image_generation` tool for that request.
 - `imageGenerationOptions`: optional `quality: "low" | "medium" | "high"`, `size: "1024x1024" | "1024x1536" | "1536x1024"`, and `background: "auto" | "opaque" | "transparent"`. Transparent requests use Codex standalone Images with `gpt-image-2`; other requests use the hosted Responses tool. Transparent mode rejects tools/structured output, supports up to five references, validates real transparent and visible PNG pixels, and never invokes background-removal fallback. Returned size/quality may differ from requested settings; read `result.providerMetadata.qgrid.imageGeneration` or per-file `result.content` metadata. `result.files[]` contains bytes/mediaType only. Logs separate `requestedOptions` from `observedGeneration`; image usage is separate from the zero text-driver usage on this route.
 
@@ -199,6 +199,12 @@ Reference images for image generation use normal AI SDK multimodal message parts
 
 ## Response mapping
 
+Cancelling the returned provider stream aborts its HTTP request and releases the SSE reader; caller abort signals propagate through the same request-local cancellation path. Normal completion and error events also release the reader. OpenAI affinity coordinates are swept for expiration on insertion and bounded to the most recently issued 1,000 entries; eviction keeps full-history generation functional.
+
+Since 2.10.0, `generateText` exposes the parent request-log ID at `result.providerMetadata?.qgrid?.requestLogId`. For `streamText`, consume the stream first, then read `(await result.providerMetadata)?.qgrid?.requestLogId`. The raw query response and stream `done` event carry optional top-level `requestLogId`.
+
+Update both server and SDK to use this metadata. The field is omitted when logging is disabled or an older server does not return it. Tool-run steps share the parent ID; `runContext.requestLogId` remains separate continuation state. This does not expose external-provider telemetry logger IDs or add IDs to stream delta/error events.
+
 qgrid response content maps to AI SDK content:
 
 - qgrid `text` -> AI SDK text content.
@@ -222,7 +228,7 @@ For a successful Fable refusal fallback (Fable 5 or 5.1):
 - `providerMetadata.qgrid.modelFallbacks` preserves the refusal route and optional category/explanation.
 - `providerMetadata.qgrid.costSource` reports whether cost came from Claude Code or qgrid's pricing table. Prefer the provider-reported combined cost for this path.
 
-Do not expose the requested Fable model as `response.modelId` after Opus served the answer, and do not map this upstream safety behavior onto the reserved `providerOptions.qgrid.fallbackModels` option.
+Do not expose the requested Fable model as `response.modelId` after Opus served the answer, and do not map this upstream safety behavior onto the deprecated `providerOptions.qgrid.fallbackModels` option.
 
 Usage maps qgrid standard usage into AI SDK V3 usage:
 

package/skills/qgrid/references/anthropic-claude-code-runtime.md

@@ -77,7 +77,7 @@ claude -p
   --model <canonical-model-or-1m-suffix>
   --system-prompt <text>                  # small system prompt
   --system-prompt-file <path>             # large system prompt
-  --thinking disabled                       # omitted for Fable 5/5.1, Opus 5, Opus 5.5 and Sonnet 5.5
+  --thinking disabled                       # omitted for Fable 5/5.1, Opus 5, Opus 5.5, Sonnet 5.5 and Haiku 5.5
   --effort <effort-or-low>
   --disable-slash-commands
   --session-id <uuid>
@@ -97,7 +97,7 @@ Important details:
 - Large system prompts over 64 KiB are written to a temporary file to avoid argv `E2BIG` — this
   same branch absorbs large injected schema contracts, so the old 64 KiB schema argv limit no
   longer applies (only the global 512 KiB caller-schema complexity limit remains).
-- `--thinking disabled`, `MAX_THINKING_TOKENS=0`, and adaptive thinking env suppression keep thinking off for existing models. Fable 5, Fable 5.1, Opus 5.5 and Sonnet 5.5 require always-on adaptive thinking (Fable: CLI catalog `rejects_disabled_thinking`; Opus 5.5: official spec, thinking cannot be turned off; Sonnet 5.5: API `{type: "disabled"}` returns 400 and the CLI catalog marks `rejects_disabled_thinking`). Opus 5 defaults to adaptive thinking and rejects disabled thinking at `xhigh`/`max` effort. qgrid omits all three suppressors for these models (`usesAdaptiveThinking`) and uses `effort` to control depth.
+- `--thinking disabled`, `MAX_THINKING_TOKENS=0`, and adaptive thinking env suppression keep thinking off for existing models. Fable 5, Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 require always-on adaptive thinking (Fable: CLI catalog `rejects_disabled_thinking`; Opus 5.5: official spec, thinking cannot be turned off; Sonnet 5.5: API `{type: "disabled"}` returns 400 and the CLI catalog marks `rejects_disabled_thinking`; Haiku 5.5 (released 2026-10-07): the API accepts disabled thinking only at `high` effort or below and returns 400 at `xhigh`/`max`, and the Claude Code 2.1.293 catalog marks `rejects_disabled_thinking`). Opus 5 defaults to adaptive thinking and rejects disabled thinking at `xhigh`/`max` effort. qgrid omits all three suppressors for these models (`usesAdaptiveThinking`) and uses `effort` to control depth.
 
 ## Spawn env
 
@@ -260,6 +260,7 @@ qgrid's exact 1M support set currently includes:
 - `claude-opus-5-5`
 - `claude-sonnet-5-5`
 - `claude-sonnet-5`
+- `claude-haiku-5-5`
 - `claude-sonnet-4-6`
 - `claude-opus-4-6`
 - `claude-opus-4-8`

package/skills/qgrid/references/request-log-run-lifecycle.md

@@ -38,6 +38,8 @@ Candidate discovery is only a hint; each candidate is rechecked and finalized in
 
 ## Tool-call loop
 
+Since 2.10.0, the frame returns the parent `requestLogId` at the top level of successful query responses and stream `done` events, including terminal text responses. It retains the allocated ID if logging postprocessing fails after provider success. The field is absent with `logger: false`. Tool-run steps share the ID, while `runContext.requestLogId` remains reserved for continuing pending tool calls. SDK callers read `providerMetadata.qgrid.requestLogId`; streaming metadata becomes available at completion. Delta and error events are unchanged.
+
 The AI SDK provider tracks pending tool-call IDs. When the next AI SDK call contains all required tool results, it sends the existing logging `runContext` when one was returned and sends the tool results. With `logger: false`, it still correlates the pending calls locally and continues the generation without a request-log id.
 
 qgrid converts tool results into continuation input for providers. Cold fallback still includes a "continue using these results" text input.
@@ -61,6 +63,8 @@ Provider-qualified ids widen `model_name` and `requested_model_name` from 50 to
 
 ## Usage fields
 
+The dashboard total includes driver `cost_usd` and `image_cost_usd` once each. Both the stored sum and legacy driver-price adjustment apply the same normalized Sonamu filters as the list. The `/api/qgrid/stats` `requests` field is an in-memory count of completed provider generations since process start, separated by provider and token name; tool follow-up generations count separately. It is not a persisted lifetime count or a count of failed attempts.
+
 Stored request log usage uses qgrid-standard semantics:
 
 - `input_tokens`: total input, including cache read/write.

package/skills/qgrid/references/sonamu-api-web-flow.md

@@ -72,7 +72,7 @@ Clicking an attached or generated image opens a preview with extension, decoded
 
 Token credential previews must tolerate missing `accessToken` rather than reading `.length` unconditionally. Legacy system-keychain credentials display `System Keychain`; this display compatibility does not enable an Antigravity runtime.
 
-Chat model presets: the model `<select>` lists the served subset of `QgridSupportedModel` grouped by provider (`MODEL_PRESET_GROUPS` in `ChatWidget.tsx`), newest first, and defaults to `anthropic/claude-fable-5-1`. Ids that the ChatGPT-subscription Codex route no longer serves (`gpt-5.4` and `gpt-5.4-mini` retired 2026-08-31 with `gpt-5.6-terra` / `gpt-5.6-luna` as replacements; `gpt-5.2` and `gpt-5.3-codex` removed earlier), `gpt-5.3-codex-spark` (rejected on the ChatGPT-account Codex route and absent from its catalog as of 2026-09-29), `gpt-5.5` (retires from the Codex/ChatGPT route on 2026-10-14; replacements `gpt-6.1-sol` / `gpt-6-luna`), the retired `claude-opus-4` / `claude-sonnet-4` (2026-06-15) and `claude-opus-4-1` (2026-08-05; Claude Code silently serves `claude-opus-5-5` instead), and the catalog-less `claude-sonnet-4-7` stay in the SDK type and price table (legacy request-log cost) but are excluded here. web does not depend on ai-sdk, so a new model must be added in both places. `openai/gpt-6-astra`, `openai/gpt-6.1-sol`, `openai/gpt-6-sol`, and `openai/gpt-6-luna` are included in the OpenAI group and `anthropic/claude-opus-5-5` and `anthropic/claude-sonnet-5-5` in the Anthropic group; adding them does not change the default model.
+Chat model presets: the model `<select>` lists the served subset of `QgridSupportedModel` grouped by provider (`MODEL_PRESET_GROUPS` in `ChatWidget.tsx`), newest first, and defaults to `anthropic/claude-fable-5-1`. Ids that the ChatGPT-subscription Codex route no longer serves (`gpt-5.4` and `gpt-5.4-mini` retired 2026-08-31 with `gpt-5.6-terra` / `gpt-5.6-luna` as replacements; `gpt-5.2` and `gpt-5.3-codex` removed earlier), `gpt-5.3-codex-spark` (rejected on the ChatGPT-account Codex route and absent from its catalog as of 2026-09-29), `gpt-5.5` (retires from the Codex/ChatGPT route on 2026-10-14; replacements `gpt-6.1-sol` / `gpt-6-luna`), the retired `claude-opus-4` / `claude-sonnet-4` (2026-06-15) and `claude-opus-4-1` (2026-08-05; Claude Code silently serves `claude-opus-5-5` instead), and the catalog-less `claude-sonnet-4-7` stay in the SDK type and price table (legacy request-log cost) but are excluded here. web does not depend on ai-sdk, so a new model must be added in both places. `openai/gpt-6-astra`, `openai/gpt-6.1-sol`, `openai/gpt-6-sol`, and `openai/gpt-6-luna` are included in the OpenAI group and `anthropic/claude-opus-5-5`, `anthropic/claude-sonnet-5-5`, and `anthropic/claude-haiku-5-5` in the Anthropic group; adding them does not change the default model.
 
 The monit tab's vitals strip shows per-provider in-flight counts, an in-flight sparkline, token chips with quota gauges (fed by the dispatcher's 60-second rate-limits cache; `resetsAt` arrives as ms epoch), and a 1-hour per-provider request/error/cache-hit stats line polled separately every 30 seconds.
 

package/skills/qgrid/references/token-auth-quota-lifecycle.md

@@ -42,6 +42,8 @@ Use this reference before changing token storage, OAuth flows, token activation,
 On creation, the database defaults `reauth_required = false`; `TokenModel.save` also applies `quota_threshold = 80` and `weight = 1` independently for any field not provided (skipped entirely when `id` is present).
 `TokenModel.save` is internal only; callers use the curated qgrid token APIs so they cannot write lifecycle fields directly.
 
+Account re-login reads, deletes, and replaces matching token rows in one explicit Puri transaction. A failed insert rolls back deletion. The oldest matching row supplies quota threshold (including null), weight, and dashboard order unless the caller explicitly supplies them; keepalive stays enabled if any replaced row enabled it. The new login still receives a new ID and resets active/reauth state. Internal full-pool reads use unlimited list mode rather than a 100-row page.
+
 Dashboard credential previews must tolerate legacy credentials without `accessToken`. System-keychain credentials, including a remaining Antigravity row, display `System Keychain` rather than crashing on `.length`. This 2.9.8 display fix neither deletes legacy rows nor enables Antigravity provider execution.
 
 OpenAI credentials:

package/skills/qgrid/references/verification-and-debugging.md

@@ -10,6 +10,12 @@ Use this reference to choose tests, smoke scripts, and first debugging targets f
 - Common error triage
 - Debugging rules
 
+## Keep Tests Lightweight
+
+Cover user-visible flows with a small real-provider E2E suite. Keep focused tests for costs, validation, cancellation, transactions, and actual regressions that a successful live request cannot prove. Remove obsolete behavior and duplicate assertions; do not delete distinct contracts just because their tests use mocks or are long. Call counts matter when they detect duplicate generation or logging, not merely an internal implementation choice.
+
+Prefer existing checks and fixtures. Do not add a generic harness, blanket coverage target, snapshot suite, or configuration-only test layer. After the relevant checks pass, repeat only for a new edit, failure, or concrete unresolved risk. Keep production fixes separate from test cleanup and never weaken a test to hide a discovered defect.
+
 ## Package Scripts
 
 The commands in this section apply only inside the qgrid source repository. In a downstream
@@ -18,13 +24,17 @@ project, use that project's configured task and tool runner.
 Root:
 
 - `mise run check`: qgrid docs consistency, oxlint, and oxfmt check for the repo.
+- `mise run test`: DB-free API tests, SDK tests, and CLI packaging tests. No provider credentials are needed; the OAuth relay test opens a local loopback listener.
 - `mise run build`: recursive package build.
 - `mise run dev`: parallel package dev servers.
 
 API package `qgrid-api`:
 
-- `mise exec -- pnpm --filter qgrid-api test`: Vitest run.
-- `mise exec -- pnpm --filter qgrid-api test:watch`: standalone watch mode.
+- `mise exec -- pnpm --filter qgrid-api test:unit`: DB-free tests with per-file isolation. Add a test path to select a file.
+- `mise exec -- pnpm --filter qgrid-api test:db`: model/transaction suites requiring a prepared local PostgreSQL test template. Missing DB is a failure, not a skip.
+- `mise exec -- pnpm --filter qgrid-api test`: both suites, unit first.
+- `mise exec -- pnpm --filter qgrid-api test:watch`: DB-free watch mode; use `test:db --watch` for DB tests.
+- `mise exec -- pnpm --filter qgrid-api test:coverage`: both suites, with separate `coverage/unit` and `coverage/db` reports.
 - `mise exec -- pnpm --filter qgrid-api build`: Sonamu build.
 - `mise exec -- pnpm --filter qgrid-api sonamu`: run Sonamu CLI.
 
@@ -83,7 +93,11 @@ Run focused Vitest files from the owning package when possible, then broaden if
 
 ## Smoke Scripts
 
-Smoke scripts assume a running qgrid server and real registered tokens. They can consume subscription quota; do not run them casually.
+Real-provider E2E requires a prepared isolated qgrid server, a registered token, and explicit opt-in. It consumes subscription quota; do not run it against production as part of routine checks. Never report mocked tests as live-provider verification.
+
+The default entry point is `packages/ai-sdk/e2e/e2e.ts` (`mise exec -- pnpm --dir packages/ai-sdk e2e`). It checks text, streaming, and tool-plus-structured output, and follows each returned `requestLogId` to the persisted parent and steps. Set `QGRID_REAL_PROVIDER_ACCEPTANCE=1`, `QGRID_URL`, and `QGRID_MODEL` explicitly. A comma-separated model list explicitly selects multiple provider runs. Each run uses a unique project label; no production logs are deleted. Exact prose, token counts, costs, and timing are not acceptance criteria.
+
+CI runs the DB-free tests plus existing package and empty-DB boot checks. Live E2E remains an explicit command; unavailable credentials or infrastructure mean unverified, not passed.
 
 OpenAI direct qgrid API:
 
@@ -109,16 +123,8 @@ AI SDK structured output:
 AI SDK tools plus structured output:
 
 - `packages/ai-sdk/e2e/tools-structured-output.ts`
-- This is the public-path release acceptance matrix: `qgrid()` HTTP/SSE,
-  `generateText`/`streamText`, OpenAI/Anthropic, direct final answers, and one
-  client-tool result followed by `Output.object`.
-- It consumes real provider quota and requires the explicit
-  `QGRID_REAL_PROVIDER_ACCEPTANCE=1` opt-in.
-- Uses `QGRID_URL`, `QGRID_ACCEPTANCE_OPENAI_MODEL`, and
-  `QGRID_ACCEPTANCE_ANTHROPIC_MODEL`. OpenAI direct-final coverage defaults to
-  10 generate plus 10 stream attempts and can be changed with
-  `QGRID_ACCEPTANCE_OPENAI_DIRECT_REPEATS_PER_MODE`. Request logs use the stable
-  project name `qgrid-ai-sdk-tools-output-acceptance`.
+- Focused regression probe for direct structured answers while tools are available, including OpenAI tuple output, streamed JSON, and streaming tool follow-up. The non-streaming tool execution happy path lives in the default E2E suite.
+- Uses the same explicit opt-in, server, and model selection as default E2E. Runs once per mode by default; set `QGRID_ACCEPTANCE_REPEATS` only for an intentional repeated probe.
 
 Image generation:
 
@@ -135,6 +141,11 @@ Other scripts under `scripts/smoke-test-*` and `scripts/debug-*` are ad hoc prob
 
 ## Common Error Triage
 
+Generated SSE handlers missing a newly added event field:
+
+- Sonamu's service generator reads runtime stream-decorator schemas. A stale API `dist` can keep the old event shape even with `sonamu sync --force`.
+- Run `mise exec -- pnpm --dir packages/api sonamu build api`, then `mise exec -- pnpm --dir packages/api sonamu sync --force`. Check the generated handler type and web typecheck; do not hand-edit generated services.
+
 `NO_OPENAI_WORKERS`:
 
 - The request path returns this only when there is no active OpenAI token candidate; there is no queue and no busy state.

package/bundle/dist/utils/providers/common/token-picker.js

@@ -1,14 +0,0 @@
-//#region src/utils/providers/common/token-picker.ts
-var RoundRobinPicker = class {
-	cursor = 0;
-	pick(tokens) {
-		if (tokens.length === 0) return null;
-		const picked = tokens[this.cursor % tokens.length];
-		this.cursor++;
-		return picked;
-	}
-};
-//#endregion
-export { RoundRobinPicker };
-
-//# sourceMappingURL=data:application/json;charset=utf-8;base64,eyJ2ZXJzaW9uIjozLCJmaWxlIjoidG9rZW4tcGlja2VyLmpzIiwibmFtZXMiOltdLCJzb3VyY2VzIjpbIi4uLy4uLy4uLy4uL3NyYy91dGlscy9wcm92aWRlcnMvY29tbW9uL3Rva2VuLXBpY2tlci50cyJdLCJzb3VyY2VzQ29udGVudCI6WyIvKipcbiAqIFRva2VuUGlja2VyIOKAlCDthqDtgbAg7ISg7YOdIOyVjOqzoOumrOymmCDsnbjthLDtjpjsnbTsiqQgKyBSb3VuZFJvYmluUGlja2VyIOq1rO2YhC5cbiAqL1xuXG5leHBvcnQgaW50ZXJmYWNlIFRva2VuUGlja2VyPFQgZXh0ZW5kcyB7IGlkOiBudW1iZXIgfT4ge1xuICBwaWNrKHRva2VuczogVFtdKTogVCB8IG51bGw7XG59XG5cbmV4cG9ydCBjbGFzcyBSb3VuZFJvYmluUGlja2VyPFQgZXh0ZW5kcyB7IGlkOiBudW1iZXIgfT4gaW1wbGVtZW50cyBUb2tlblBpY2tlcjxUPiB7XG4gIHByaXZhdGUgY3Vyc29yID0gMDtcblxuICBwaWNrKHRva2VuczogVFtdKTogVCB8IG51bGwge1xuICAgIGlmICh0b2tlbnMubGVuZ3RoID09PSAwKSByZXR1cm4gbnVsbDtcbiAgICBjb25zdCBwaWNrZWQgPSB0b2tlbnNbdGhpcy5jdXJzb3IgJSB0b2tlbnMubGVuZ3RoXSE7XG4gICAgdGhpcy5jdXJzb3IrKztcbiAgICByZXR1cm4gcGlja2VkO1xuICB9XG59XG4iXSwibWFwcGluZ3MiOiI7QUFRQSxJQUFhLG1CQUFiLE1BQWtGO0NBQ2hGLFNBQWlCO0NBRWpCLEtBQUssUUFBdUI7RUFDMUIsSUFBSSxPQUFPLFdBQVcsR0FBRyxPQUFPO0VBQ2hDLE1BQU0sU0FBUyxPQUFPLEtBQUssU0FBUyxPQUFPO0VBQzNDLEtBQUs7RUFDTCxPQUFPO0NBQ1Q7QUFDRiJ9

package/bundle/src/utils/providers/common/token-picker.ts

@@ -1,18 +0,0 @@
-/**
- * TokenPicker — 토큰 선택 알고리즘 인터페이스 + RoundRobinPicker 구현.
- */
-
-export interface TokenPicker<T extends { id: number }> {
-  pick(tokens: T[]): T | null;
-}
-
-export class RoundRobinPicker<T extends { id: number }> implements TokenPicker<T> {
-  private cursor = 0;
-
-  pick(tokens: T[]): T | null {
-    if (tokens.length === 0) return null;
-    const picked = tokens[this.cursor % tokens.length]!;
-    this.cursor++;
-    return picked;
-  }
-}