fix(providers): Copilot mixed-wire routing, DeepSeek service_tier + reasoning replay, Claude 4.6/4.7 1M windows by lidge-jun · Pull Request #889 · lidge-jun/opencodex

docs(plan): design the Codex Set prompt composer against upstream evidence

lidge-jun

committed

39abf4c

docs(plan): research API-side client-integration toggles (cc-switch reference)

lidge-jun

committed

a47a167

docs(plan): rebuild the prompt-composer design around its audit blockers

lidge-jun

committed

91b3921

docs(plan): drop the TOML dependency after measuring Bun's parser

lidge-jun

committed

921070b

docs(plan): stop journal recovery from overwriting other people's edits

lidge-jun

committed

3195c71

docs(plan): allocate 12 must-fix bugs across 5 bugfix worktrees

lidge-jun

committed

ef71180

docs(plan): make the journal one state machine and finish the WP1 contract

lidge-jun

committed

c327fdb

docs(plan): fold audit blockers into bugfix worktree units

lidge-jun

committed

470618d

docs(plan): reclassify Plugins after reading the OR it actually depends on

lidge-jun

committed

14eae93

docs(plan): close three of the four NEAR-PASS residuals before building

lidge-jun

committed

072b8df

docs(plan): design the unified Integrations surface with guaranteed rollback

lidge-jun

committed

60b417c

feat(codex): prompt-layer inventory and a TOML encoder we can trust

lidge-jun

committed

e0b15fe

docs(plan): use ~ in example paths so privacy:scan stays green

lidge-jun

committed

478354e

docs(plan): split wt3 roadmap into per-bug decade docs with researched mechanisms

lidge-jun

committed

47e7d61

docs(plan): fold audit round-1 blockers into wt3 bug-a docs

lidge-jun

committed

2bb3d91

docs(plan): fold audit round-2 blockers — sol in built-in set, additive seed policy

lidge-jun

committed

6a9b79e

docs(plan): fold audit round-3 — uniform seven-model count, honest precedence fixture

lidge-jun

committed

62bbeb5

fix(providers): route Copilot Responses-only models off chat completions

lidge-jun

committed

14c8214

docs(providers): Copilot Responses-only GPT-5 defaults + modelAdapters escape hatch

lidge-jun

committed

e67c0df

docs(plan): wp-b P — concrete file map and reviewed service_tier semantics

lidge-jun

committed

cbd345b

docs(plan): fold wp-b audit round-1 blockers

lidge-jun

committed

198a317

docs(plan): fold wp-b audit round-2 residual — fail-closed scope wording in 000

lidge-jun

committed

b09596e

fix(responses): gate service_tier by provider capability, preserve DeepSeek reasoning replay

lidge-jun

committed

b42d573

docs(providers): supportsServiceTier + preserveResponsesReasoningContent reference rows, capability-gated fast-tier guides

lidge-jun

committed

1e3c050

fix(providers): keep service-tier/reasoning capabilities registry-only

lidge-jun

committed

9f801e3

docs(plan): wp-c P — stale check narrows scope to registry map + agents-inject + generator override

lidge-jun

committed

a0ea408

docs(plan): fold wp-c audit round-1 — generated metadata + pinned tests + full regression set

lidge-jun

committed

2843eb5

fix(providers): give Claude 4.6/4.7 their 1M context windows; authoritative [1m] in generated profiles

lidge-jun

committed

0b30283

Navigation Menu

fix(providers): Copilot mixed-wire routing, DeepSeek service_tier + reasoning replay, Claude 4.6/4.7 1M windows - #889

fix(providers): Copilot mixed-wire routing, DeepSeek service_tier + reasoning replay, Claude 4.6/4.7 1M windows#889
lidge-jun wants to merge 28 commits into
devfrom
codex/wt3-exec

Commits on Aug 2, 2026

docs(plan): design the Codex Set prompt composer against upstream evidence

docs(plan): research API-side client-integration toggles (cc-switch reference)

docs(plan): rebuild the prompt-composer design around its audit blockers

docs(plan): drop the TOML dependency after measuring Bun's parser

docs(plan): stop journal recovery from overwriting other people's edits

docs(plan): allocate 12 must-fix bugs across 5 bugfix worktrees

docs(plan): make the journal one state machine and finish the WP1 contract

docs(plan): fold audit blockers into bugfix worktree units

docs(plan): reclassify Plugins after reading the OR it actually depends on

docs(plan): close three of the four NEAR-PASS residuals before building

docs(plan): design the unified Integrations surface with guaranteed rollback

feat(codex): prompt-layer inventory and a TOML encoder we can trust

docs(plan): use ~ in example paths so privacy:scan stays green

docs(plan): split wt3 roadmap into per-bug decade docs with researched mechanisms

docs(plan): fold audit round-1 blockers into wt3 bug-a docs

docs(plan): fold audit round-2 blockers — sol in built-in set, additive seed policy

docs(plan): fold audit round-3 — uniform seven-model count, honest precedence fixture

fix(providers): route Copilot Responses-only models off chat completions

docs(providers): Copilot Responses-only GPT-5 defaults + modelAdapters escape hatch

docs(plan): wp-b P — concrete file map and reviewed service_tier semantics

docs(plan): fold wp-b audit round-1 blockers

docs(plan): fold wp-b audit round-2 residual — fail-closed scope wording in 000

fix(responses): gate service_tier by provider capability, preserve DeepSeek reasoning replay

docs(providers): supportsServiceTier + preserveResponsesReasoningContent reference rows, capability-gated fast-tier guides

fix(providers): keep service-tier/reasoning capabilities registry-only

docs(plan): wp-c P — stale check narrows scope to registry map + agents-inject + generator override

docs(plan): fold wp-c audit round-1 — generated metadata + pinned tests + full regression set

fix(providers): give Claude 4.6/4.7 their 1M context windows; authoritative [1m] in generated profiles