Skip to content

fix(router-provider): fetch model metadata before context management decisions - #1053

Merged
edelauna merged 6 commits into
mainfrom
fix/router-provider-context-window
Jul 31, 2026
Merged

fix(router-provider): fetch model metadata before context management decisions#1053
edelauna merged 6 commits into
mainfrom
fix/router-provider-context-window

Conversation

@JamesRobert20

@JamesRobert20 JamesRobert20 commented Jul 29, 2026

Copy link
Copy Markdown
Contributor

Problem

Router providers that are auth-scoped (zoo-gateway, kimi-code) skip the model cache entirely. When a fresh handler is constructed (on provider switch or profile change), getModel() falls back to hardcoded default model info because the real model list has not been fetched from the API yet.

Context management in Task.ts calls getModel() before createMessage() (where fetchModel() normally runs). This means condensing and truncation decisions use the default context window (200k for zoo-gateway) instead of the actual model's context window.

If the user selects a model with a larger context window through the gateway (e.g. Opus 4.8 at 1M tokens), the system incorrectly calculates context usage against 200k, triggering premature condensation.

Fix

  • Add ensureModelFetched() to RouterProvider that calls fetchModel() once when the instance model map is empty. Subsequent calls are no-ops.
  • Add ensureModelFetched as an optional method on the ApiHandler interface so Task.ts can call it.
  • Call await this.api.ensureModelFetched?.() in Task.ts before both context management paths (attemptApiRequest and handleContextWindowExceededError).

Static providers (Anthropic, Gemini, Bedrock, etc.) are unaffected since they don't implement the method. Non-auth-scoped router providers (OpenRouter, Vercel AI Gateway) already have cache hits via getModelsFromCache() so the fetch is skipped after the first call.

Summary by CodeRabbit

  • New Features
    • Model metadata (context window and pricing) can now be prefetched automatically via an optional API capability.
    • Router-based model loading is now concurrency-safe and memoized to avoid duplicate requests.
  • Bug Fixes
    • Context-limit handling and token management now use freshly fetched model metadata instead of defaults.
    • If metadata fetching fails, requests continue using safe fallback behavior.
    • Updated model details (e.g., context window) correctly override prior default values.
  • Tests
    • Expanded coverage for cache short-circuiting, recovery after failures, and concurrency/deduplication.

…decisions

Router providers (zoo-gateway, kimi-code) that are auth-scoped skip the
model cache entirely. On a fresh handler instance getModel() falls back
to hardcoded defaults (e.g. 200k context window) because the real model
list has not been fetched yet. Context management runs before
createMessage() which is where fetchModel() normally happens, so
condensing/truncation decisions use the wrong context window.

Add ensureModelFetched() to RouterProvider that fetches once when the
instance model map is empty. Call it in Task before context management
so getModel() returns accurate metadata from the API.

Co-authored-by: Cursor <cursoragent@cursor.com>
@coderabbitai

coderabbitai Bot commented Jul 29, 2026

Copy link
Copy Markdown
Contributor

Review Change Stack

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: cd0bc5b3-b80b-4eb8-859c-328fc0d600c6

📥 Commits

Reviewing files that changed from the base of the PR and between c7ead3e and 1264874.

📒 Files selected for processing (1)
  • src/core/task/__tests__/Task.spec.ts

📝 Walkthrough

Walkthrough

ApiHandler now optionally supports lazy model fetching. RouterProvider caches and deduplicates remote model metadata requests, while task flows safely fetch metadata before using model-derived values.

Changes

Model Metadata Fetching

Layer / File(s) Summary
Provider model fetching
src/api/index.ts, src/api/providers/router-provider.ts, src/api/providers/__tests__/zoo-gateway.spec.ts
The API contract and RouterProvider add memoized model fetching with concurrent-call deduplication, retry recovery, and tests for cache behavior and fetched context windows.
Task model-dependent flows
src/core/task/Task.ts, src/core/task/__tests__/Task.spec.ts
Task request and context-window paths use a failure-tolerant model-fetch helper, with tests covering invocation, missing handlers, logging, and continued streaming.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Task
  participant ApiHandler
  participant RouterProvider
  Task->>ApiHandler: ensureModelFetched()
  ApiHandler->>RouterProvider: ensureModelFetched()
  RouterProvider->>RouterProvider: fetchModel() when models are empty
  RouterProvider-->>Task: model metadata available
Loading

Possibly related PRs

  • Zoo-Code-Org/Zoo-Code#344: Adds the Zoo Gateway model discovery and fetching behavior covered by the updated provider tests.

Suggested labels: awaiting-review

Suggested reviewers: edelauna

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name Status Explanation Resolution
Description check ⚠️ Warning The description explains the problem and fix, but it omits required template sections like the linked issue, test procedure, and checklist. Add the required sections from the template: Related GitHub Issue, Test Procedure, Pre-Submission Checklist, and any applicable notes.
✅ Passed checks (4 passed)
Check name Status Explanation
Title check ✅ Passed The title clearly summarizes the main change: prefetching router model metadata before context decisions.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.
✨ Finishing Touches
📝 Generate docstrings
  • Create stacked PR
  • Commit on current branch
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Commit unit tests in branch fix/router-provider-context-window

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@codecov

codecov Bot commented Jul 29, 2026

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 93.75000% with 1 line in your changes missing coverage. Please review.

Files with missing lines Patch % Lines
src/core/task/Task.ts 85.71% 0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

@coderabbitai coderabbitai Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/api/providers/router-provider.ts`:
- Around line 64-68: Make ensureModelFetched() single-flight by storing and
reusing an in-flight fetchModel() promise, clearing it after completion while
preserving the empty-model check. Update ZooGatewayHandler.createMessage() and
completePrompt() to call ensureModelFetched() followed by getModel(), replacing
their unconditional fetchModel() calls so auth-scoped providers avoid duplicate
metadata requests.

In `@src/core/task/Task.ts`:
- Around line 4046-4047: Move the optional ensureModelFetched call before the
cachedStreamingModel snapshot and before getSystemPrompt-dependent request
setup. Reuse the initialized model returned by this.api.getModel() for both
cachedStreamingModel and modelInfo, ensuring fresh router providers have
metadata available even when contextTokens is falsy.
🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

  • Push a commit to this branch (recommended)
  • Create a new PR with the fixes

ℹ️ Review info
⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: bc932c48-2b90-434b-b92a-9aef4da2165e

📥 Commits

Reviewing files that changed from the base of the PR and between dcaa3cb and 101b490.

📒 Files selected for processing (4)
  • src/api/index.ts
  • src/api/providers/__tests__/zoo-gateway.spec.ts
  • src/api/providers/router-provider.ts
  • src/core/task/Task.ts

Comment thread src/api/providers/router-provider.ts
Comment thread src/core/task/Task.ts Outdated
…ll site

Make ensureModelFetched single-flight so concurrent callers share a
single in-flight fetch instead of firing duplicates. Move the call site
before the cachedStreamingModel snapshot so the model info is accurate
from the start of the streaming session, not just for context management.

Co-authored-by: Cursor <cursoragent@cursor.com>

@coderabbitai coderabbitai Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

♻️ Duplicate comments (1)
src/core/task/Task.ts (1)

4048-4049: 🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Move model fetching before getSystemPrompt() and remove the contextTokens guard.

This is the same still-valid issue from the previous review. Direct attemptApiRequest() callers compute getSystemPrompt() before fetching metadata, and requests with falsy contextTokens skip fetching entirely, leaving router-provider defaults in model-dependent logic. Fetch at the start of attemptApiRequest() before getSystemPrompt().

🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/core/task/Task.ts` around lines 4048 - 4049, Update attemptApiRequest to
call ensureModelFetched at its start, before getSystemPrompt is computed, and
remove the contextTokens guard that currently allows fetching to be skipped.
Preserve the existing model metadata usage through getModel().info after the
unconditional fetch.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Duplicate comments:
In `@src/core/task/Task.ts`:
- Around line 4048-4049: Update attemptApiRequest to call ensureModelFetched at
its start, before getSystemPrompt is computed, and remove the contextTokens
guard that currently allows fetching to be skipped. Preserve the existing model
metadata usage through getModel().info after the unconditional fetch.

ℹ️ Review info
⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 0d25401b-9da0-459f-9f39-79608468aa3a

📥 Commits

Reviewing files that changed from the base of the PR and between 101b490 and 47e0732.

📒 Files selected for processing (3)
  • src/api/providers/__tests__/zoo-gateway.spec.ts
  • src/api/providers/router-provider.ts
  • src/core/task/Task.ts
🚧 Files skipped from review as they are similar to previous changes (2)
  • src/api/providers/tests/zoo-gateway.spec.ts
  • src/api/providers/router-provider.ts

@github-actions github-actions Bot added the awaiting-review PR changes are ready and waiting for maintainer re-review label Jul 29, 2026
Comment thread src/core/task/Task.ts Outdated

await this.diffViewProvider.reset()

await this.api.ensureModelFetched?.()

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The litellm/deepseek/moonshot fetchers re-throw, so a failure here would reject into the catch at L3743, which return trues and ends the task silently — the comment there assumes only attemptApiRequest can throw. Pre-change, the same failure surfaced via the createMessage retry path. Would catching and logging here be safer, letting getModel() fall back to defaults?

Comment thread src/api/providers/router-provider.ts Outdated

private modelFetchPromise?: Promise<void>

async ensureModelFetched(): Promise<void> {

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For auth-scoped providers, does the first request now fetch the model list twice — once here and again unconditionally in createMessage (zoo-gateway.ts L184)? getModels bypasses both caches for auth-scoped providers (modelCache.ts L265), so the second call is another network round trip. Worth making fetchModel() single-flight (or short-circuiting when this.models is populated) so both callers share one fetch?

})
})

describe("ensureModelFetched", () => {

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is the rejection path covered anywhere? If getModels rejects once and the finally cleanup in router-provider.ts (L71-74) ever regresses, the stored rejected promise would poison every subsequent call. A reject-then-resolve test here would lock that in.

@github-actions github-actions Bot added awaiting-author PR is waiting for the author to address requested changes and removed awaiting-review PR changes are ready and waiting for maintainer re-review labels Jul 29, 2026
…ouble-fetch

Make fetchModel single-flight and short-circuit once models are loaded so
auth-scoped providers do not hit the models endpoint twice per request.
Catch ensureModelFetched failures in Task via safeEnsureModelFetched so a
metadata fetch error falls back to defaults instead of ending the task.
Add reject-then-recover coverage and Task tests for the new call sites.

Co-authored-by: Cursor <cursoragent@cursor.com>
@github-actions github-actions Bot removed the awaiting-author PR is waiting for the author to address requested changes label Jul 30, 2026
JamesRobert20 and others added 2 commits July 30, 2026 05:10
vi.spyOn on the private method fails check-types; route it through the
existing test access cast like the other private helpers.

Co-authored-by: Cursor <cursoragent@cursor.com>
@github-actions github-actions Bot added the awaiting-author PR is waiting for the author to address requested changes label Jul 30, 2026
@github-actions github-actions Bot added awaiting-author PR is waiting for the author to address requested changes and removed awaiting-author PR is waiting for the author to address requested changes labels Jul 31, 2026
@edelauna
edelauna added this pull request to the merge queue Jul 31, 2026
@github-actions github-actions Bot added awaiting-review PR changes are ready and waiting for maintainer re-review and removed awaiting-author PR is waiting for the author to address requested changes labels Jul 31, 2026
Merged via the queue into main with commit ca9b60f Jul 31, 2026
15 checks passed
@edelauna
edelauna deleted the fix/router-provider-context-window branch July 31, 2026 16:07
xavier-arosemena added a commit to xavier-arosemena/roo-plus that referenced this pull request Aug 3, 2026
…tocol, slim ClineProvider, upstream sync (#112)

* refactor(webview): use canonical provider identifiers (Zoo-Code-Org#1023)

* refactor(webview): use canonical provider identifiers

* test(webview): remove brittle identifier mutation tests

* fix(webview): handle remaining model providers explicitly

* test(webview): avoid unreachable coverage branch

* refactor(webview): use canonical anthropic identifier

* test(webview): cover Anthropic Opus 1M tier

* test(webview): use canonical Anthropic identifier

* fix(webview): address canonical provider review

* test(webview): use canonical provider identifiers

* test(webview): cover Kimi Code model selection

* refactor: finish canonical provider identifier audit (Zoo-Code-Org#1030)

* refactor: finish canonical provider identifier audit

* fix: address provider identifier review feedback

* refactor: centralize Anthropic protocol value

* refactor: reuse canonical provider protocol constants

* refactor: centralize OpenAI protocol value

* test: use canonical keyless provider identifiers

* test: address protocol routing review feedback

* refactor(api): centralize service-tier primitives (Zoo-Code-Org#1040)

* refactor(api): centralize service-tier primitives

* test(openai-native): cover omitted completion tiers

* test(openai-native): cover resolved streaming tiers

* test(api): cover service tier fallbacks

* test(api): cover remaining service tier branches

* refactor(api): remove duplicate tier capture

* test(openai-native): cover flex service tier

* test(openai-native): expand service tier coverage

* refactor: address service tier review feedback

* refactor(api): address service tier review feedback

* test(webview): add model info visual snapshot

* fix(webview): isolate visual test dependencies

* test(webview): update service tier visual baseline

* fix(webview): add @source directive so Tailwind utilities generate in CT builds

---------

Co-authored-by: Elliott de Launay <edelauna@gmail.com>

* fix(openai-compatible): add max reasoning effort option (Zoo-Code-Org#882) (Zoo-Code-Org#1051)

* feat(openai-codex): persist and send Fast priority mode (Zoo-Code-Org#1063)

* feat(openai-codex): persist and send Fast priority mode

* refactor(openai-codex): align service tier types

* fix(readiness): remove stale cloud test, fix floating promises, rebrand webview title

* chore: bump custom-modes submodule (user-friendly descriptions) + add roomodes sync guard

- Point custom-modes at 029a0d4 (description rewrite + YAML normalization)
- Make sync-custom-modes.mjs importable and add scripts/verify-roomodes-sync.mjs
  which fails when .roomodes is not reproducible from the submodule

* feat(protocol): add zod message registry + boundary validation (S1-M1/M2)

* feat(protocol): type + validate allowedCommands/deniedCommands (S1-M3 domain 1)

- Add commands.ts schemas registered in webviewMessageSchemas (allowedCommands, deniedCommands require a string[]).
- Migrate the handler cases to schema-validated typed payloads, dropping the runtime Array.isArray/typeof sanitization now guaranteed by zod.
- Boundary now rejects crafted non-array commands payloads (unit + handler + ClineProvider boundary tests).

* feat(protocol): type + validate updateSettings (S1-M3 domain 2)

- rooCodeSettingsSchema now uses .passthrough() so unknown future settings fields are retained, not stripped.
- Add settings.ts updateSettings schema (updatedSettings?: RooCodeSettings) registered in webviewMessageSchemas.
- Migrate the updateSettings handler case to the schema-validated payload; malformed known-field types (e.g. non-string terminalProfile) are rejected before side effects.
- Boundary + handler tests cover valid dispatch and malformed rejection.

* feat(protocol): type + validate provider config messages (S1-M3 domain 3)

- Add providerConfig.ts schemas for saveApiConfiguration, upsertApiConfiguration (text + ProviderSettings-passthrough) and setApiConfigPassword, registered in webviewMessageSchemas.
- apiConfiguration reuses providerSettingsSchema.passthrough(): key fields (incl. apiProvider enum) validated, provider-specific passthrough fields retained.
- Migrate save/upsert handler cases to schema-validated typed payloads, dropping the runtime text/apiConfiguration guards.
- setApiConfigPassword remains a no-op but malformed shapes are now rejected at the boundary.

* feat(protocol): type + validate marketplace install messages (S1-M3 domain 4)

- Add marketplace.ts schemas for installMarketplaceItem, installMarketplaceItems (min 1 item) and installMarketplaceItemWithParameters, registered in webviewMessageSchemas.
- mpItem/mpItems reuse marketplaceItemSchema; mpInstallOptions reuses installMarketplaceItemOptionsSchema.
- Migrate the three handler cases to schema-validated typed payloads; crafted non-item payloads are rejected at the boundary.

* feat(protocol): type + validate chat message queue messages (S1-M3 domain 5)

- Add messageQueue.ts schemas for queueMessage (text + images), removeQueuedMessage (text) and editQueuedMessage (payload reused from queuedMessageSchema.pick), registered in webviewMessageSchemas.
- Migrate the three handler cases to schema-validated typed payloads, removing the payload-as-EditQueuedMessagePayload cast and the now-unused import.
- Boundary + handler tests cover valid dispatch and malformed rejection (non-string text).

* feat(protocol): type + validate todo list and custom mode messages (S1-M3 domain 6)

- Add customModes.ts schemas for updateTodoList (payload.todos from todoItemSchema), updateCustomMode (slug + modeConfig) and deleteCustomMode (slug + optional checkOnly), registered in webviewMessageSchemas.
- Migrate the three handler cases to schema-validated typed payloads; updateTodoList drops the payload-as-any cast (handler no-explicit-any suppression 5 -> 4).
- Fix the ClineProvider updateCustomMode boundary test to include the top-level slug, matching the real webview sender contract (do not loosen the schema).
- Boundary + handler tests cover valid dispatch and malformed rejection.

* test(protocol): guard updateSettings sender compatibility

Add parseWebviewMessage assertions for the CLI extension-host initialSettings
shape and the webview SettingsView handleSubmit payload (incl. nullable/edge
fields) so the updateSettings schema never regresses the real senders.

* ci: add message-schema ratchet guard (S1-M4)

* refactor(webview): extract shared handler helpers and narrow provider types (S2 scaffold)

- Add handlers/shared.ts with getGlobalState/updateGlobalState/getCurrentCwd/resolveIncomingImages
  extracted verbatim from webviewMessageHandler's pre-switch setup.
- Narrow provider params in checkpointRestoreHandler, generateSystemPrompt, worktree/handlers,
  skillsMessageHandler, rulesMessageHandler to minimal Pick<ClineProvider, ...> types so domain
  handlers depend only on the members they use (dependency inversion; ClineProvider satisfies
  each Pick structurally, so callers need no cast).
- Exempt core/webview/handlers/*.ts from no-case-declarations (same rule the original dispatcher
  used for its switch case bodies).
- Re-home the 4 no-explicit-any suppressions from webviewMessageHandler.ts to handlers/chat.ts
  (the moved casts keep their existing suppressions).

No behavior change: moved helper bodies are identical and the old dispatcher still compiles
against the narrowed signatures.

* refactor(webview): add per-domain webview message handler modules (S2)

Move every case from webviewMessageHandler's giant switch into domain modules under
core/webview/handlers/, each exporting a ReadonlySet<WebviewMessageType> of the types it
handles plus a handle<Domain>Messages(provider, marketplaceManager, message) function that
switch-dispatches those cases VERBATIM (same bodies, ordering, and error handling).

  chat.ts            17 types (message edit/delete/confirm, queue, tts, checkpoints, enhancePrompt)
  task.ts            17 types (new/clear/cancel/condense/export/delete tasks, system prompt, commits, todos)
  settings.ts        27 types (updateSettings, allowed/deniedCommands, custom modes, prompts, models)
  providerProfiles.ts 15 types (api config CRUD, pins, provider OAuth sign-in/out, rate limits)
  mcp.ts              9 types (server lifecycle, tool toggles, timeout, settings)
  marketplace.ts      7 types (install/remove/filter/fetch, mdm notification)
  worktree.ts        11 types (list/create/delete/switch worktrees, branches, includes, picker)
  codeIndex.ts        8 types (settings, indexing lifecycle, secrets, auto-enable)
  skills.ts           6 types   rules.ts 5 types   commands.ts 4 types
  terminal.ts         3 types   images.ts 3 types   debug.ts   3 types
  misc.ts            19 types (webviewDidLaunch, import/export, files, search, upsells, preview)

Each module declares a minimal Pick<ClineProvider, ...> limited to the members it actually uses.
The old dispatcher is untouched in this commit; the router that consumes these modules lands next.

No behavior change; cases are byte-for-byte the original statements relocated.

* refactor(webview): replace dispatcher switch with thin domain router (S2)

webviewMessageHandler.ts shrinks from a 4k-line switch to a ~170-line router:
it builds a Map<WebviewMessageType, handler> from each domain module's exported
MessageTypes set + handle<Domain>Messages function and delegates by message.type.
The exported signature (provider, message, marketplaceManager?) is unchanged, so the
boundary (ClineProvider.setWebviewMessageListener) and all spec files pass untouched.
Unknown types fall through to the same commented default as before.

The 4 no-explicit-any suppressions that moved to handlers/chat.ts in the scaffold commit
are now the only re-homed entry; webviewMessageHandler.ts is fully typed (0 any).

* refactor(webview): extract TaskHistoryService from ClineProvider (S3a)

Move task-history mutation, webview broadcast, debounced globalState
write-through, and recent-tasks caching into a focused TaskHistoryService
with narrow DI ports. ClineProvider keeps identical public method
signatures and delegates to the service. recentTasksCache stays on the
provider (owned through a port) so delegation flows and existing tests
that read/write the field keep working unchanged.

* refactor(webview): extract ProviderProfileService from ClineProvider (S3a)

Move provider-profile CRUD, activation, and sticky-profile persistence into
a focused ProviderProfileService with narrow DI ports. ClineProvider keeps
identical public method signatures and delegates. providerSettingsManager
is injected as a getter port (read at call time) so tests that replace the
field after construction keep working; updateTaskHistory is forwarded with
exact argument arity to preserve spy call signatures.

* refactor(webview): extract MarketplaceService from ClineProvider (S3a)

Move on-demand marketplace data fetching into a focused MarketplaceService
with narrow DI ports. The timeout warning is injected as a port (wired to
vscode.window.showWarningMessage in ClineProvider) so the service has no
direct vscode dependency and stays unit-testable. ClineProvider keeps the
same public fetchMarketplaceData signature and delegates.

* feat(core): extract TaskOrchestrator service for task lifecycle and delegation state machine

Adds src/core/services/TaskOrchestrator.ts owning the task lifecycle and
delegation/subtask state machine previously embedded in ClineProvider (S3b).

- TaskOrchestratorDeps: narrow DI ports (S3a pattern) bound to the provider at
  call time so spies (getState/getGlobalState/updateTaskHistory/getTaskWithId)
  and post-construction taskRegistry/taskScheduler swaps keep working.
- Moves createTask, createTaskWithHistoryItem, cancelTask/cancelTaskInternal,
  clearTask, resumeTask, addClineToStack, removeClineFromStack, evictCurrentTask,
  markDelegatedChildInterrupted, delegateParentAndOpenChild,
  reopenParentFromDelegation, abandonSubtask, and runDelegationTransition.
- Behavior preserved byte-for-byte (error messages, assertValidTransition order,
  instanceId guards, cancelledDelegationChildIds semantics, tool_result injection,
  flush/retry, log output).
- Adds focused unit tests at the narrowest layer (single-open invariant on
  createTask, cancelTask instanceId guard + rehydrate, delegation parent-metadata
  persistence + child scheduling, reopenParentFromDelegation tool_result injection,
  abandonSubtask orphan/transition logic).

* refactor(core): slim ClineProvider by delegating task orchestration to TaskOrchestrator

ClineProvider now keeps only webview lifecycle, state assembly, settings/misc,
and thin delegates. Every moved method remains a public method on ClineProvider
with an identical name/signature/return type that delegates to the orchestrator
via a lazily-cached static helper (ClineProvider.getTaskOrchestrator(this)), so:

- Delegation specs that invoke ClineProvider.prototype.<method>.call(fakeProvider)
  against plain `this` objects keep working (no reliance on the prototype chain).
- vi.spyOn(provider, getState|getGlobalState|updateTaskHistory|getTaskWithId) and
  post-construction provider.taskRegistry/taskScheduler swaps keep intercepting
  (all deps are closures read at call time).
- handleModeSwitch/showTaskWithId/performPreparationTasks/getTaskWithId stay on the
  provider and are consumed via narrow ports.

Moved methods: addClineToStack, removeClineFromStack, evictCurrentTask,
markDelegatedChildInterrupted, createTaskWithHistoryItem, createTask, cancelTask,
cancelTaskInternal, clearTask, resumeTask, delegateParentAndOpenChild,
reopenParentFromDelegation, abandonSubtask, runDelegationTransition.

Also removes now-unused imports and drops the ClineProvider eslint-suppression
count from 12 to 6 (no increase; no new suppressions).

* X1: add DOMPurify sanitizeHtml primitive for webview dangerouslySetInnerHTML sites

- Add dompurify dependency to webview-ui and a strict allowlist sanitizer
  (sanitizeHtml.ts) using an isolated DOMPurify instance (no global hooks).
- TerminalOutput: keep escapeXML pinned to true (exported const + regression
  test) and sanitize converter output as belt-and-suspenders.
- MermaidBlock: set mermaid securityLevel to strict (was loose) so labels are
  HTML-escaped at the source.
- MermaidButton: sanitize the copied SVG HTML before injecting into the zoom
  modal (child-node rendering would detach the live diagram).
- TaskItem: sanitize search-highlight HTML before injecting.
- Add sanitizeHtml.spec.ts asserting script/event-handler/javascript: removal,
  span+color preservation, and SVG path allow-listing with nested script
  stripping.

* X2: tighten HMR CSP and verify the local dev server is Vite before serving HMR HTML

- Remove the https://* wildcard from the dev-only HMR CSP (script-src,
  style-src, connect-src). script-src now allows only https://*.posthog.com
  (telemetry), the local Vite origins, and the nonce. 'unsafe-eval' is kept
  (required by Vite HMR/react-refresh) with a comment that it is dev-only and
  must never appear in getHtmlContent's production CSP.
- Harden the localhost probe: after the root reachability check, GET
  /@vite/client and require a 2xx whose body identifies Vite, otherwise fall
  back to getHtmlContent (production HTML). A rogue process on :5173 can no
  longer serve scripts to the webview. Keeps the .vite-port gate and the
  existing hmr_not_running error message.
- Extend ClineProvider.spec.ts with a dev-mode getHMRHtmlContent suite
  (scoped beforeEach/afterEach reset/restore the axios mock since the outer
  beforeEach's vi.clearAllMocks does not clear mockImplementationOnce queues):
  CSP has no bare https://* wildcard, and the vite-identity probe falls back to
  production HTML when /@vite/client is unreachable or not-Vite.

* docs: update changelog, debt log, readme and add typed-message-protocol ADR

* [Fix] Subtask e2e suite can inherit a cancelled delayed mock stream from the previous test (Zoo-Code-Org#1074)

* test(e2e): drain delayed mock stream deterministically in subtask suite

The API-hang subtask fixture used aimock's flat latency, which applies
per SSE chunk and is never interrupted by client disconnects, so a
cancelled delayed stream stayed pending server-side for chunks x latency
and could flush into the next test's traffic.

- delay only the first chunk via streamingProfile.ttft so the pending
  window is exactly the shared SUBTASK_API_HANG_RESPONSE_LATENCY_MS
- anchor the post-test drain to the request's aimock journal timestamp
  and wait out only the remainder of that bounded window

* docs(e2e): clarify subtask drain invariants from code review

---------

Co-authored-by: Roomote <roomote@roomote.dev>
Co-authored-by: Elliott de Launay <edelauna@gmail.com>

* fix(router-provider): fetch model metadata before context management decisions (Zoo-Code-Org#1053)

* fix(router-provider): fetch model metadata before context management decisions

Router providers (zoo-gateway, kimi-code) that are auth-scoped skip the
model cache entirely. On a fresh handler instance getModel() falls back
to hardcoded defaults (e.g. 200k context window) because the real model
list has not been fetched yet. Context management runs before
createMessage() which is where fetchModel() normally happens, so
condensing/truncation decisions use the wrong context window.

Add ensureModelFetched() to RouterProvider that fetches once when the
instance model map is empty. Call it in Task before context management
so getModel() returns accurate metadata from the API.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(router-provider): single-flight ensureModelFetched and earlier call site

Make ensureModelFetched single-flight so concurrent callers share a
single in-flight fetch instead of firing duplicates. Move the call site
before the cachedStreamingModel snapshot so the model info is accurate
from the start of the streaming session, not just for context management.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(router-provider): address review feedback on fetch failures and double-fetch

Make fetchModel single-flight and short-circuit once models are loaded so
auth-scoped providers do not hit the models endpoint twice per request.
Catch ensureModelFetched failures in Task via safeEnsureModelFetched so a
metadata fetch error falls back to defaults instead of ending the task.
Add reject-then-recover coverage and Task tests for the new call sites.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(test): spy private addToApiConversationHistory via TaskTestAccess

vi.spyOn on the private method fails check-types; route it through the
existing test access cast like the other private helpers.

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>

* v3.76.0: Release Readiness & Architecture Program — typed message protocol, slim ClineProvider, upstream sync

- Merge upstream/main (7 commits: canonical provider identifiers, router-provider metadata, openai-codex Fast mode, service-tier refactor, e2e subtask fix)
- Preserve Zoo Gateway removal from the fork; drop upstream's zoo-gateway references/tests that conflict with removed provider
- Typed + runtime-validated webview message protocol (16 security-sensitive types)
- Domain-split webview dispatcher and slim ClineProvider (TaskOrchestrator, TaskHistoryService, ProviderProfileService, MarketplaceService)
- Semble one-dir EACCES fix, download-only binary, pre-installed mode description merge-fill
- CLI event-listener leak fix, legacy credential write path retired, vscode-shim logger wired
- Webview HTML sanitization (DOMPurify) + HMR CSP hardening
- Fix path-mentions double-escaping regression; remove blank Zoo Gateway test stubs
- Update CHANGELOG.md + README.md; bump version to 3.76.0

Closes: #98

Co-authored-by: hanneke-de-vries <dhanneke204@gmail.com>

* fix: remediate new CodeQL alerts (port validation, temp files, command race, path escaping)

Co-authored-by: hanneke-de-vries <dhanneke204@gmail.com>

* fix: remediate 100 pre-existing CodeQL code-scanning alerts

- js/file-system-race: drop fs.access pre-checks in favor of direct
  read-with-error-handling (McpHub, extract-text, ReadFileTool), use
  readdirSync withFileTypes (find-missing-i18n-key), and exclusive 'wx'
  writes (bootstrap.mjs)
- js/insecure-temporary-file: use fs.mkdtemp private dirs + 0600 modes
  (diagnosticsHandler, ShadowCheckpointService spec/service)
- js/remote-property-injection: validate property keys in ModesView and
  FileChangesPanel
- js/disabling-certificate-validation: expand justification comment for
  the debug-only TLS override (networkProxy)
- js/file-access-to-http / http-to-file-access: sanitize image
  references (image-generation), validate base64 image payloads
  (openai-native, openai-codex), validate image bytes before writing
  (GenerateImageTool), validate OAuth token response schema (qwen-code)
- js/indirect-command-line-injection: use spawnSync with args array +
  editor allowlist (install-vsix)
- js/log-injection: sanitize control chars in mock-server URL logs
- js/missing-origin-check: add isTrustedMessage origin/source validator
  and apply it across ~30 webview message handlers

Co-authored-by: hanneke-de-vries <dhanneke204@gmail.com>

* fix: remediate remaining CodeQL alerts flagged on the branch

- ReadFileTool.ts: open a single file handle and stat/read through it to
  eliminate the stat-then-read (TOCTOU) race in both the native and
  legacy read paths
- FileChangesPanel.tsx + ModesView.tsx: store webview-sourced content by
  Map key instead of object property to prevent prototype pollution
  (remote-property-injection) via untrusted paths/slugs
- mcp-oauth.test.ts: sanitize the HTTP method as well as the URL in the
  mock-server log (log-injection)
- qwen-code.ts: validate refresh_token as a string primitive before
  persisting credentials (network-data-to-file)
- apps/vscode-e2e/tsconfig.json: use non-deprecated moduleResolution
  (Node10 + ignoreDeprecations 5.0) and set explicit rootDir
- readFileTool.spec.ts: mock fs.open handle so tests exercise the new
  single-handle read path (no new eslint suppressions)

Co-authored-by: hanneke-de-vries <dhanneke204@gmail.com>

---------

Co-authored-by: Alexei Gubin <36731953+WebMad@users.noreply.github.com>
Co-authored-by: Elliott de Launay <edelauna@gmail.com>
Co-authored-by: Ivan Ramadhan Arifin <111653938+ivanarifin@users.noreply.github.com>
Co-authored-by: zoomote[bot] <305051434+zoomote[bot]@users.noreply.github.com>
Co-authored-by: Roomote <roomote@roomote.dev>
Co-authored-by: James Mtendamema <59908268+JamesRobert20@users.noreply.github.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: hanneke-de-vries <dhanneke204@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

awaiting-review PR changes are ready and waiting for maintainer re-review

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants