Repository navigation
0.1.8 Beta6: Complete the DeepSeek V4 Flash 0731 Chat CLI path #396
Copy link
Copy link
Closed
Labels
P1Priority 1: current iteration goalPriority 1: current iteration goalcampaign:0.1.8-beta60.1.8 Beta6 usable Chat campaign0.1.8 Beta6 usable Chat campaigncampaign:gateway-chatGateway deep modules and Codex↔Chat bridgeGateway deep modules and Codex↔Chat bridgeenhancementNew feature or requestNew feature or requestready-for-agentFully specified, ready for an AFK agentFully specified, ready for an AFK agenttier:heavyOrchestrator provider tier: heavyOrchestrator provider tier: heavywayfinder:taskWayfinder implementation taskWayfinder implementation task
Description
Parent
Outcome
Run the real Codex CLI through the complete Beta6 text, standard-tool, and Collaboration V2 contracts on the Ollama Cloud deepseek-v4-flash:0731 Chat Completions endpoint.
This evidence is independent from GLM-5.2 and K2.7 Code and must use the same generic conversion implementation.
Acceptance criteria
- Progressive text, multi-turn history, default reasoning, and a standard function call/result lifecycle complete through Ollama Chat.
- A real read -> target-scoped apply_patch -> verify loop completes.
- Collaboration V2 completes spawn, canonical task identity/status, message, follow-up, wait/result, list/interrupt, child result delivery, parent completion, streaming/history, and same-Home restart/readback.
- Runtime fields and V2 items are preserved, consumed, reversibly adapted, or rejected by value according to 0.1.8 Beta6: Generate the runtime-derived Chat and Collaboration V2 conversion matrix #66.
- Exact Provider/model identity remains
deepseek-v4-flash:0731; no Responses/Official fallback or cross-Provider execution occurs. - Slow-valid streams, upstream availability, incomplete SSE, client cancellation, and semantic incompatibility remain distinct.
- No DeepSeek-specific production conversion branch is introduced.
- Evidence is bounded and candidate/CLI/Provider/model/protocol bound.
Blocked by
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
P1Priority 1: current iteration goalPriority 1: current iteration goalcampaign:0.1.8-beta60.1.8 Beta6 usable Chat campaign0.1.8 Beta6 usable Chat campaigncampaign:gateway-chatGateway deep modules and Codex↔Chat bridgeGateway deep modules and Codex↔Chat bridgeenhancementNew feature or requestNew feature or requestready-for-agentFully specified, ready for an AFK agentFully specified, ready for an AFK agenttier:heavyOrchestrator provider tier: heavyOrchestrator provider tier: heavywayfinder:taskWayfinder implementation taskWayfinder implementation task
Activity
1 remaining item
NOirBRight commentedon Aug 19, 2026
Authenticated DeepSeek V4 Flash 0731 Chat qualification passed on source candidate
45e06184f00b5ced9df40069c8fede0423a98840; bounded evidence is in4d7807cunderdocs/evidence/authenticated-provider/chat/summary.json.Real
codex-cli 0.147.0observations: non-stream Chat 200, progressive CLI text and reasoning, exact exec -> target-scoped apply_patch -> verify, all six Collaboration V2 tools, child result, same-Home restart/readback, Cross-Home rejection before Gateway mutation, exact ollama-cloud/deepseek-v4-flash:0731 identity, fallback=0, V1=false. Deliberate child interruption remains visible as bounded downstream 499 rather than Provider success; observed 400/transport attempts remain explicit. No DeepSeek-specific production branch was added.Verification: focused 357 passed + 16 subtests; full Python core 2,458 passed / 90 skipped / 508 subtests, with only 61 known Linux PowerShell/Windows host gaps; quality parse_errors=0.
NOirBRight commentedon Aug 19, 2026
Final authenticated qualification supersedes the earlier preliminary evidence.
Source candidate:
73e4ed3d444414dedcc1a5902fa9f0a9c848d469Evidence commit:
c82baad(docs/evidence/authenticated-provider/chat/summary.json)DeepSeek V4 Flash 0731 passed the real
codex-cli 0.147.0 -> Gateway Responses -> ollama-cloud /v1/chat/completions -> Responsespath with: non-stream 200; 11 progressive text streams; same-session text resume; explicit reasoning policy; exact standard/custom call-result identity; target-scoped apply_patch; independently qualified six-tool V2 phases; canonical task, frozen result shapes, child delivery and history identity; same-Home restart/readback; Cross-Home rejection before request. Exact Provider/model/protocol remained bound, fallback=0, V1=false. Request/semantic 400 and deliberate client-cancellation 499 remain separately categorized; slow-valid/incomplete/availability distinctions remain protocol-controlled by #395 fixtures. No DeepSeek-specific production branch was added.Final checks: 366 focused tests + 16 subtests; full Python core 2,458 passed / 90 skipped / 508 subtests with only the 61 known Linux PowerShell/Windows-host gaps; quality parse_errors=0. Final Standards/Spec review found no Spec blocker.
test(gateway): qualify authenticated Provider Chat routes (#285 #286 #…
docs(evidence): record three-model authenticated Chat qualification (#…
test(evidence): require real Chat reasoning observations (#285 #286 #396
test(gateway): deepen authenticated Chat lifecycle qualification (#285 …
test(gateway): accept valid V2 message follow-up ordering (#285 #286 #…
test(gateway): qualify V2 tools in isolated session turns (#285 #286 #…
test(gateway): scope Provider qualifier to Chat issues (#285 #286 #396)
test(gateway): recover bounded V2 child delivery waits (#285 #286 #396)
docs(evidence): deepen three-model Chat qualification proof (#285 #286 …