* Expand extended cache TTL model list + add message-breakpoint sub-toggle
Two related changes to Anthropic Messages API prompt caching:
1. Expand modelSupportsExtendedCacheTtl beyond just the 1M context
variants. Per Anthropic docs, the 1h cache TTL is available on all
active models; this widens our opt-in list to Claude Opus 4.5/4.6/4.7
and Sonnet 4.5/4.6 (all variants, not just -1m).
2. Add a new experiment-based setting
chat.anthropic.promptCaching.extendedTtlMessages as a strict
sub-toggle of the existing extendedTtl setting. When both are on, the
rolling message-level breakpoints (last cacheable user / tool-result
blocks set by addMessagesApiCacheControl) also use the 1h TTL
instead of the default 5m. Nested rather than orthogonal because
Anthropic requires longer-TTL breakpoints to appear before shorter
ones in the tools->system->messages prefix order.
Tests: 69 passed. Added a suite for isExtendedCacheTtlMessagesEnabled
(parent on/off x sub on/off matrix + inherited model/location/subagent
gates) and two tests verifying addMessagesApiCacheControl propagates
the new cacheTtl argument.
* Slim down extended cache TTL tests
- Trim isExtendedCacheTtlEnabled model-list test to just verify the
delegation (full boundaries are covered by modelSupportsExtendedCacheTtl).
- Remove redundant 'inherits gates from parent' test in
isExtendedCacheTtlMessagesEnabled suite — the parent×sub matrix plus
the parent's own gate tests already cover this.
- Merge the two addMessagesApiCacheControl ttl tests into one
parameterized assertion.
* Update stale comment about message breakpoint TTL
The comment claimed message breakpoints 'always use the default 5m TTL',
but that's no longer true when the new extendedTtlMessages sub-toggle is on.
* Refactor extended cache TTL: pass parentEnabled, drop misleading coercions
- isExtendedCacheTtlMessagesEnabled now takes parentEnabled: boolean
instead of re-running the parent gate. Call site passes the resolved
useExtendedCacheTtl directly, eliminating a duplicate experiment-service
lookup per request. Makes the 'sub-toggle of' relationship literal in
the signature.
- Drop the !! coercion on getExperimentBasedConfig<boolean> returns —
the generic guarantees T, so the coercion was misleading defensive
noise.
- Narrow cacheTtl param from '5m' | '1h' to just '1h' on both
addToolsAndSystemCacheControl and addMessagesApiCacheControl. Per
Anthropic docs, { type: 'ephemeral' } already defaults to 5m, so '5m'
is never actually emitted on the wire and call sites never passed it.
- Stronger composition test for isExtendedCacheTtlEnabled — replaces
four single-axis tests with one table-driven matrix that exercises
all four gates simultaneously, catching short-circuit refactors.
- Table-driven 2x2 matrix for isExtendedCacheTtlMessagesEnabled.
- Trim user-facing extendedTtlMessages setting description; team-only
rationale (rolling breakpoints, 2x write premium) lives in the JSDoc.
- Update stale comment claiming message breakpoints always use 5m.
* Remove outdated comments about extended cache TTL models
Thinking blocks are signed/encrypted by the producing model and tied to its API surface. The previous gate short-circuited the same-model check for the current turn ('!isHistorical || ...'), so a round produced by a different model in the same turn — e.g. a subagent handoff or a mid-turn model switch — would still have its thinking replayed through the endpoint's model. Anthropic Messages rejects this with a signature validation error; Responses treats it as garbage reasoning state.
Make sameModelAsEndpoint a hard precondition for including thinking, regardless of whether the round is historical, matching the existing convention used a few lines below for phase.
Related: https://github.com/microsoft/vscode/issues/316536
- Use `frontmatter` (one word)
- Align `Show source` icon and text with the reopen with
- Make the `Open preview` in explorer use `reopen with` instead so you can nicely toggle the opened preview
The override resolver's non-Copilot vendor branch (which goes through
vscode.lm.selectChatModels) was not exercised by the existing tests.
Add four cases stubbing lm.selectChatModels with sinon: success returns
an ExtensionContributedChatEndpoint; zero matches, multiple matches
(ambiguous), and rejection all fall back to the default copilot-utility.
Addresses PRRT_kwDOAn8RLM6CZMCp.
Validate the override value at runtime: getNonExtensionConfig<string> is a TS cast only, so a user putting a non-string in settings.json (e.g. 42) would crash getChatEndpoint with raw.indexOf is not a function. Treat any non-string value as 'no override' and ignore.
Subscribe to vscode.lm.onDidChangeChatModels and forward it as a model refresh. The copilot model fetcher's own onDidModelsRefresh covers the copilot vendor; this catches the BYOK case where the underlying provider can refresh or remove models out-of-band.
Add a regression test for the non-string override path.
isUserSelectable: false does not stop an extension from resolving `vscode.lm.selectChatModels({ vendor: 'copilot', id: 'copilot-utility[-small]' })` and using the returned LanguageModelChat to call sendRequest. When the alias resolved to a BYOK override, that proxied requests to the user's configured BYOK endpoint without going through that provider's authorization flow.
Skip publishing the alias entry under the copilot vendor when the resolved endpoint is not produced by this provider. The internal _utilityAliasEndpoints map is still populated, and copilot's own internal callers continue to honor the override via IEndpointProvider.getChatEndpoint(family) — they don't depend on the alias being published through the public LM API surface.
* Add archived state management and worktree sharing for chat sessions
- Introduced `archived` property in `ChatSessionMetadataFile` to track archived sessions.
- Updated `IChatSessionMetadataStore` interface to include methods for setting and getting archived state.
- Implemented logic in `ChatSessionMetadataStore` and `MockChatSessionMetadataStore` to handle archived sessions.
- Added `getBlockingSiblingSessionsForFolder` utility to identify sessions that share worktrees.
- Modified CLI chat session commands to respect archived state during session deletion and worktree cleanup.
- Updated tests to cover new functionality and ensure proper behavior of session management.
* Add worktree sharing edge-case unit tests
Co-authored-by: DonJayamanne <1948812+DonJayamanne@users.noreply.github.com>
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Extract the alias-info builder into a pure helper and pass an explicit isCopilotProviderEndpoint flag, so a BYOK model whose id collides with a copilot model id (e.g. 'gpt-4o') is no longer misidentified as copilot and cloned with the wrong metadata.
Subtract baseCount + BaseTokensPerCompletion from the synthesized maxInputTokens so the reported context window matches the convention used by the regular model entries (instead of overstating it).
Add unit tests for the helper covering the clone, BYOK id-collision, and missing-base cases.
SemanticSearchTextSearchProvider caches the copilot-utility-small endpoint after the first use. Without subscribing to IEndpointProvider.onDidModelsRefresh, changing chat.utilitySmallModel later did not affect the search flow until the provider was recreated. Make the provider Disposable, subscribe to the refresh event, and dispose the provider alongside its registration.
Reject ambiguous overrides (more than one matching model) instead of silently picking the first result, which could route requests to the wrong configured endpoint.
Demote per-resolution info logs to trace; utility endpoints are used by many background flows and unguarded info logs are noisy when an override is configured.
Restrict the chat.utilityModelOverride telemetry event to successful applications only, and drop the vendor/modelId fields. Those values can be user-provided for BYOK/custom endpoint providers and should not be reported. Clear the per-family fingerprint when the relevant setting changes so a new value emits once.
Flips the default for `chat.freezeCustomizationsIndex` from `false` to `true` so the customizations payload (`<instructions>`/`<skills>`/`<agents>`) is frozen at the first turn of a conversation and reused on subsequent turns, stabilizing the system-prompt prefix for cache reuse.
Loosen `_registerUtilityAliasModels` so the copilot-utility / copilot-utility-small aliases are still published when the resolved endpoint belongs to a non-copilot vendor (e.g. a BYOK Anthropic / Gemini model selected via chat.utilityModel).
Previously, the alias was published only if `endpoint.model` matched
a model already in this providers `models` list — a copilot-vendor
model. Any consumer asking for `vscode.lm.selectChatModels({ vendor:
copilot, id: copilot-utility })` would then get nothing and the
override would be silently unreachable for that codepath.
The new behavior:
- copilot-vendor endpoint: clone the existing entry (preserving
rich provider metadata like tooltip, pricing, requiresAuthorization,
etc.).
- non-copilot endpoint: synthesize a minimal entry from the
endpoint itself (name, family, version, max input/output tokens,
supportsToolCalls, supportsVision). The alias is hidden from the
picker (`isUserSelectable: false`); the missing rich fields are not
user-visible. The vendor is `copilot` either way because this
provider is registered under that vendor.
Add trace logs that distinguish the cloned vs synthesized branches so
the alias mapping is visible in logs.
When the user has set chat.utilityModel / chat.utilitySmallModel, route the corresponding internal utility family through the configured model instead of the default CAPI selection.
The override is read in `_resolveUtilityFamily` and resolved by a new `_resolveUtilityOverride` helper:
- The setting value is encoded as `${vendor}/${id}` (matching the
workbench storage format). It is parsed once and routed by vendor:
- copilot vendor: looked up via `_modelFetcher.getAllChatModels()`.
Going via `lm.selectChatModels({vendor: copilot, ...})` from
inside this codepath would re-enter the language-model
services per-vendor `_resolveLMSequencer` (which is held while
the copilot LM provider is preparing its model list — this same
resolution path runs as part of utility-alias publishing). That
re-entrancy deadlocks the picker. The fast path through
`_modelFetcher` avoids it.
- other vendors: looked up via `lm.selectChatModels` and wrapped
in `ExtensionContributedChatEndpoint` for actual request
dispatch. Different vendors use different sequencer keys, so
no deadlock.
- Any failure (malformed value, no match, lookup throws, fetch
throws) silently falls back to the existing
`CopilotUtility{Small,}ChatEndpoint.resolve()` path. Failures are
logged at warn level so they show up in default logs without
spamming.
- A constructor-side `onDidChangeConfiguration` subscription fires
`_onDidModelsRefresh` when either setting changes so the alias
endpoints are reresolved on the next request.
Also wires telemetry (`chat.utilityModelOverride`) for adoption and outcome tracking. The event is deduped per family via a fingerprint of `${raw}|${outcome}|${vendor}|${modelId}` so we emit at most once per genuine change to the setting or to which model it resolves to. Only successful applications (`applied-copilot` / `applied-extension`) include vendor + modelId; failure outcomes leave them blank.
/search/repositories does not include private repos unless the query is scoped via user:/org: qualifiers. Append the authenticated user's login and their org memberships to the search query so private repos accessible to the permissive token surface in the agents repo picker.
Fixes#309659
Introduces dedicated endpoint resolver classes (CopilotUtilitySmallChatEndpoint,
CopilotUtilityChatEndpoint) in place of the ModelAliasRegistry indirection, and
republishes the resolved utility endpoints under their new family ids
('copilot-utility-small', 'copilot-utility') as LanguageModelChatInformation
entries so that workbench callers using selectLanguageModels({ vendor: 'copilot',
id: 'copilot-utility-small' }) keep working.
- Renames the internal family identifiers everywhere they're consumed:
callers, tests, and workbench code in src/vs/workbench/contrib/chat/.
- Drops src/platform/endpoint/common/modelAliasRegistry.ts.
- CopilotUtilitySmallChatEndpoint.resolve tries a small primary model and
falls back to a second small model if the primary is unavailable.
- CopilotUtilityChatEndpoint.resolve returns the API-marked default base
model (is_chat_fallback === true), preserving existing behavior.
- Updates _copilotBaseModel field and 'copilot-base' literal in
ModelMetadataFetcher to _copilotUtilityModel / 'copilot-utility'.
No user-facing behavior change. The setting-driven 'disabled'/'default'/BYOK
selector support will be layered on top in a follow-up.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
remove comment
remove code smell
remove more code smell
* tool call events for AH and Claude CLI as well
* consistent ownership
* share definitions
* comment
* Remove Agent Host telemetry changes (split to aamunger/ToolCallTelemetry-AH)
copilot-swe-agent[bot]
Bhavya U
Vijay Upadya
Vritant Bhardwaj
Matt Bierner
guomaggie
Vikram Nitin
dileepyavan
Don Jayamanne
Paul
Osvaldo Ortega
Logan Ramos
Aaron Munger
Dmitriy Vasyura
Connor Peet