Skip to content

feat: openai-compatible-strict-reasoning (2/2) - #34

Closed
myk1yt wants to merge 53 commits into
pr/b05a-strict-reasoning-v2from
pr/b17-provider-cost-v2
Closed

feat: openai-compatible-strict-reasoning (2/2)#34
myk1yt wants to merge 53 commits into
pr/b05a-strict-reasoning-v2from
pr/b17-provider-cost-v2

Conversation

@myk1yt

@myk1yt myk1yt commented Aug 2, 2026 •
edited
Loading

Copy link
Copy Markdown
Owner

Stack Position

Description

Full Feature Description

  • Feature Branch: feat/openai-compatible-strict-reasoning
  • Feature Name: OpenAI-Compatible Strict Reasoning and Provider Cost
  • Purpose: Resolves the problem where, despite differing levels of strict JSON schema and reasoning effort support across OpenAI-compatible endpoints, request control is scattered per provider, and cost is calculated as zero or inaccurately due to usage field differences. Enables users to explicitly opt in to strict tool schema and extended reasoning effort on supported endpoints, and normalizes response usage according to provider contracts to consistently display actual cost.
  • Full Change Description: B05a adds provider settings contracts, cached settings UI, strict schema opt-in, reasoning effort values, and base request shaping. B17 normalizes conditional fields, cached tokens, missing/zero values, and price lookup for OpenAI, OpenAI-compatible, Anthropic Vertex, and Qwen family usage. B05a is also shared as the foundation for the MiMo chain, but in this chain B17 is a sibling outcome that does not depend on B12.
  • Impact Scope: Affects provider-settings.ts, base-openai-compatible-provider.ts, base-provider.ts, OpenAICompatible.tsx, openai.ts, openai-compatible.ts, anthropic-vertex.ts, qwen-code.ts.
  • Errors and Edge Cases: Strict mode is opt-in and does not change existing schemas when disabled. Strict/reasoning fields are not sent to unsupported providers. MCP schemas are not aggressively hardened. The settings UI only modifies cachedState before saving. Cost calculation treats missing fields as unknown or zero per provider contract and does not produce negative tokens. Cached input/output tokens and provider-specific price units are not double-counted. B17 does not change request payload or tool-call policy.
  • Testing Method: Run B05a's settings serialization, schema conversion, opt-in/omitted-field, request fixture, and UI/locale tests, and B17's provider-specific usage/cost fixtures. Manually compare requests with strict toggle off vs. on, and feed fixed usage fixtures into OpenAI-compatible and non-OpenAI providers, comparing against documented calculation formulas and costs.

Why Split Into 17 PRs

Instead of submitting this feature as a single unified PR, it was split into individual PRs because as code size grows, safely reviewing a PR becomes very difficult. The feature was broken into mutually exclusive individual PRs so that each can be reviewed independently.

What This PR Specifically Changes

Normalizes OpenAI/OpenAI-compatible/Anthropic Vertex/Qwen usage fields and cached tokens, and calculates cost according to provider price lookup. Does not change request payload, strict UI, or MiMo tool policy.

Included Files

  • src/api/providers/openai.ts
  • src/api/providers/openai-compatible.ts
  • src/api/providers/anthropic-vertex.ts
  • src/api/providers/qwen-code.ts
  • Direct usage/cost tests for each provider

Exclusion Scope

  • Request execution method changes
  • MiMo tool-call policy
  • B05a strict-reasoning settings/UI
  • Unrelated provider fixture changes
  • All items in the common removal rules

@myk1yt myk1yt changed the title B17: Provider Cost v2 feat: openai-compatible-strict-reasoning (2/2) Aug 3, 2026
@myk1yt
myk1yt force-pushed the pr/b17-provider-cost-v2 branch from ca96c3c to 7f5fa9e Compare August 4, 2026 06:11
@myk1yt
myk1yt force-pushed the pr/b05a-strict-reasoning-v2 branch from 658c4f2 to 4692197 Compare August 4, 2026 11:34
@myk1yt
myk1yt force-pushed the pr/b17-provider-cost-v2 branch from 7f5fa9e to 42df2fa Compare August 4, 2026 11:40
@myk1yt
myk1yt force-pushed the pr/b05a-strict-reasoning-v2 branch 3 times, most recently from 77e7207 to a3b22a7 Compare August 4, 2026 20:40
@myk1yt
myk1yt force-pushed the pr/b17-provider-cost-v2 branch 5 times, most recently from 98dd23f to 9d2c50a Compare August 6, 2026 07:28
Zoo (VP) and others added 18 commits August 7, 2026 04:11
Zoo (VP)
feat: add model-level tool-call capability and policy resolution
76d3837
Zoo (VP)
feat: wire MiMo provider controls and tighten argument normalization
c290d77 
# Conflicts:
#	src/core/tools/error-interception/StructuralValidator.ts
Zoo (VP)
feat: add ghost quarantine and max-one tool call enforcement
b989fdd 
# Conflicts:
#	src/core/assistant-message/__tests__/NativeToolCallParser.spec.ts
#	src/core/assistant-message/__tests__/presentAssistantMessage-parser-dedup.integration.spec.ts
#	src/core/assistant-message/presentAssistantMessage.ts
Zoo (VP)
feat: add tool-call policy telemetry events
7fba2f0 
# Conflicts:
#	src/core/assistant-message/__tests__/presentAssistantMessage-parser-dedup.integration.spec.ts
Zoo (VP)
fix: resolve no-explicit-any lint errors in mimo and telemetry files
307baa4
Zoo (VP)
fix: preserve parallel behavior for known providers without explicit …
45b99c7 
…capabilities
Zoo (VP)
fix: port NativeToolParseFailure infrastructure and clean error-inter…
47deb83 
…ception refs from backup
Zoo (VP)
fix: port cleaned mimo.ts provider from backup to match spec types
5f21220
Zoo (VP)
fix(mimo): suppress parallel tool calls at provider stream level
e07abd9
@myk1yt Zoo (VP)
fix(mimo): apply strict tool schemas via convertToolsForOpenAI()
1a9dfc2 
MimoHandler was passing raw tool schemas to the API without the
strict mode conversion that all other OpenAI-compatible providers use.
This caused tool call errors due to missing required/strict fields.

- Call this.convertToolsForOpenAI(tools) instead of raw assignment
- Adds strict: true, required properties, additionalProperties: false
Zoo (VP)
fix(mimo): pass openAiToolStrictMode setting to convertToolsForOpenAI
3cd9242
Zoo (VP)
fix(mimo): pass tools to convertToolsForOpenAI without extra strictMo…
a43d4d9 
…de arg
Zoo (VP)
fix(mimo): drop argument fragments of disguised parallel tool calls
4245e3f 
An id-less argument-continuation chunk belongs to the most recent id chunk
seen at its index. When a provider reuses index 0 with a NEW id (a disguised
second parallel call), the new call's id chunk was dropped but its id-less
argument fragments were still kept and concatenated into the FIRST call's
accumulator, corrupting its JSON.

Track dropped indexes in filterToFirstToolCall state and drop subsequent
id-less fragments for those indexes. Also rewrite the function docblock,
which referenced a non-existent error-interception retry loop.
Zoo (VP)
fix: correct misleading error-interception comments in tool-call parser
280da35 
The parseErrors/parseFailures docblocks claimed presentAssistantMessage
routes recorded failures to an INVALID_JSON_ARGUMENTS error-interception
pattern. No such routing exists on this codebase; describe the actual
lifecycle (consumed via the consume* APIs, cleared on new API request).
Comment-only change, no behavior difference.
Zoo (VP)
fix: clear stale native tool-call parse failures on new API request
8267250 
parseErrors/parseFailures static maps accumulated an entry per malformed
tool call and were never cleared in production (the consume* APIs have no
production callers), slowly leaking for the extension-host lifetime.

Add NativeToolCallParser.clearParseFailures() and call it in
Task.recursivelyMakeClineRequests alongside clearAllStreamingToolCalls()/
clearRawChunkState(), where other per-stream state is reset. The consume*
APIs keep working for tests.
Zoo (VP)
fix(mimo): retry once without strict tool schemas on endpoint rejection
30256cb 
MiMo sends tools through convertToolsForOpenAI(), which attaches a strict
flag to every function tool. An OpenAI-compatible endpoint that doesn't
support structured outputs rejects the request with a 400 and the turn
fails outright.

Mirror the existing parallel_tool_calls fallback: detect schema-rejection
errors narrowly (400 status plus a mention of strict/additionalProperties
in a tools context, so unrelated 400s like MiMo's missing-reasoning_content
rejection are not retried) and retry once with the original schemas and no
strict flag.
Zoo (VP)
test(b12): add mimo error retry and ghost quarantine coverage for cod…
2a6842e 
…ecov/patch
Zoo (VP)
fix(test): replace 'as any' with typed assertion in mistral.spec.ts
b68d77b
@myk1yt myk1yt closed this Aug 7, 2026
@myk1yt
myk1yt deleted the pr/b17-provider-cost-v2 branch August 7, 2026 13:06
@myk1yt
myk1yt restored the pr/b17-provider-cost-v2 branch August 7, 2026 13:32
@myk1yt myk1yt reopened this Aug 8, 2026
Zoo (VP) and others added 21 commits August 8, 2026 14:50
Zoo (VP)
fix(test): add aimock fixture for provider-cost e2e (PR Zoo-Code-Org#…
17d2c14 
…1132)

CI failure: E2E Tests (Mocked) failed with '404 No fixture matched' because
provider-cost.test.ts calls startNewTask with probe tag 'provider-cost-e2e'
but no fixture existed.
Zoo (VP)
fix(types,e2e): allow any valid URL for mimoBaseUrl to support local …
712bb66 
…stubs and custom endpoints
Zoo (VP)
fix(vscode-e2e): emit attempt_completion in provider-cost stub and fi…
59fcd0e 
…x partial check
@edelauna
fix(telemetry): record tool usage once centrally, sanitize raw tool n…
200d2aa 
…ames (Zoo-Code-Org#1073)

* fix(telemetry): record tool usage once centrally, sanitize raw tool names

* fix(telemetry): defer native MCP usage recording until validation passes

* fix(telemetry): narrow UseMcpToolTool callback, harden test mocks, close coverage gaps

* test(telemetry): complete native MCP mock so validateToolExists runs the real path
@zoomote @roomote
refactor: reuse code-index reset helpers (Zoo-Code-Org#1194)
fb57aeb 
Co-authored-by: Roomote <roomote@roomote.dev>
@zoomote @roomote
refactor: reuse shared config test helpers (Zoo-Code-Org#1195)
d5085a2 
Co-authored-by: Roomote <roomote@roomote.dev>
@zoomote @roomote
refactor: reuse terminal test reset helpers (Zoo-Code-Org#1197)
5afa9b3 
Co-authored-by: Roomote <roomote@roomote.dev>
@zoomote @roomote
refactor: reuse webview chat render helper (Zoo-Code-Org#1196)
84e54cf 
Co-authored-by: Roomote <roomote@roomote.dev>
@edelauna @navedmerchant
fix(task): skip saveClineMessages when history task aborts before mes…
7f07a1d 
…sages load (Zoo-Code-Org#1181)

* fix(task): skip saveClineMessages when history task aborts before messages load

* test: strengthen resume-eviction-race assertions and type mock provider

* test: add fallback mock for second getSavedClineMessages read in resume-eviction spec

* fix(task): prevent saving unhydrated history messages during abort

---------

Co-authored-by: Naved  Merchant <naved.merchant@gmail.com>
@zoomote @roomote
refactor: reuse shared test helpers in config import/export spec (Zoo…
ca86e38 
…-Code-Org#1198)

Co-authored-by: Roomote <roomote@roomote.dev>
@WebMad @edelauna
refactor(webview): complete provider identifier migration (Zoo-Code-O…
a800299 
…rg#1141)

* refactor(webview): complete provider identifier migration

* test(webview): type selected model hook mocks

* test(webview): keep selected model spec out of provider migration

* test(webview): harden provider identifier migration tests

---------

Co-authored-by: Elliott de Launay <edelauna@gmail.com>
@renovate @edelauna
chore(deps): update dependency mermaid to v11.16.1 [security] (Zoo-Co…
ef32d7e 
…de-Org#1193)

* chore(deps): update dependency mermaid to v11.16.1 [security]

* fix(webview): harden mermaid securityLevel and tighten dev-mode CSP

---------

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
Co-authored-by: Elliott de Launay <edelauna@gmail.com>
@renovate
chore(deps): update dependency undici to v6.28.0 [security] (Zoo-Code…
2b2641b 
…-Org#1161)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
@zoomote @roomote
refactor: reuse shared reset helper in code-index specs (Zoo-Code-Org…
2a62b64 
…#1199)

Co-authored-by: Roomote <roomote@roomote.dev>
@zoomote @roomote
refactor: reuse shared reset helper in remaining code-index specs (Zo…
25fa8d4 
…o-Code-Org#1200)

Co-authored-by: Roomote <roomote@roomote.dev>
@WebMad @edelauna
refactor(webview): use provider identifiers in ApiOptions (Zoo-Code-O…
1c60627 
…rg#1146)

* refactor(webview): canonicalize ApiOptions provider identifiers

* test(webview): strengthen ApiOptions interaction coverage

* fix(webview): use providerIdentifiers.openrouter in ApiOptions option sort

---------

Co-authored-by: Elliott de Launay <edelauna@gmail.com>
@zoomote @roomote
refactor: reuse webview render helper in chat and settings specs (Zoo…
02f7902 
…-Code-Org#1203)

Co-authored-by: Roomote <roomote@roomote.dev>
@zoomote @roomote
refactor: reuse shared reset helper in provider specs (Zoo-Code-Org#1202
20dd966 
)

Co-authored-by: Roomote <roomote@roomote.dev>
@zoomote @roomote
refactor: reuse webview render helper in remaining settings specs (Zo…
03db526 
…o-Code-Org#1204)

Co-authored-by: Roomote <roomote@roomote.dev>
@zoomote @roomote
refactor: reuse shared reset helper in semble and terminal specs (Zoo…
4c42f91 
…-Code-Org#1201)

Co-authored-by: Roomote <roomote@roomote.dev>
@myk1yt
Merge branch 'main' into pr/b17-provider-cost-v2
9a921cf
@myk1yt

myk1yt commented Aug 10, 2026

Copy link
Copy Markdown
Owner Author

Closing to recreate with main as target base branch. This PR had stale base branch references after fork sync.

@myk1yt myk1yt closed this Aug 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@myk1yt @WebMad @edelauna