feat(agent-core-v2): round-trip reasoning_details in chat completions - #3492
Conversation
|
commit: |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 714ef33b53
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| const reasoningDetails = extractReasoningDetails(message); | ||
| if (reasoningDetails !== undefined) { | ||
| yield* convertReasoningDetails(reasoningDetails); |
There was a problem hiding this comment.
When reasoningKey is explicitly configured and a response also contains an array-valued reasoning_details, this branch bypasses ReasoningKeyDialect.observe and accepts the array anyway. The resulting detailsIndex parts are subsequently serialized as reasoning_details plus reasoning_content, rather than under the configured key, so the explicit setting no longer pins the dialect as its existing contract and test describe; apply details detection only when no explicit key is set, in both response paths.
Useful? React with 👍 / 👎.
714ef33 to
38ec6e0
Compare
38ec6e0 to
3c63038
Compare
Sync 147 upstream commits onto the fork. Notable upstream changes: - Remove the protocol package and relocate wire types into agent-core-v2/src/events.ts (MoonshotAI#3646) - Absorb stepRetry into the LLM requester turn state machine (already followed in the 2026-09-07 sync) - NotifyUser tool + mid-turn update panel (MoonshotAI#3524) - HEIC/HEIF/BMP image support (MoonshotAI#3649), resumable large file reads (MoonshotAI#3645), reasoning_details round-trip (MoonshotAI#3492), zstd updater manifests (MoonshotAI#3669) Conflict resolutions (all by hand, fork features preserved): - package.json: drop the empty simple-git-hooks block (upstream cleanup; not a fork feature). - docs/config-manifest.toml: take the union, then regenerate with pnpm gen:config-manifest (33 sections, fork sections intact). - test/agent/stepRetry/stepRetry.test.ts: keep upstream's slimmed retryBackoffDelays test — the fork had deleted the old fat version when stepRetry was absorbed; the util is fork-used code with no other coverage. - protocol/src/events.ts: accept upstream's package removal; the fork's compaction.started model/model_display fields already auto-merged into their new home (agent/fullCompaction/compactionOps.ts). No source still imports @moonshot-ai/protocol. Merge follow-ups: - Regenerate state/wire/config manifests; freshness tests pass. - node-sdk update-all-session-models flag test: drop the v1-engine parity half (legacy agent-core v1 is removed upstream); keep the v2 registration pin. - Fix pre-existing lint errors in merge-touched files so the lint-staged pre-commit gate passes: unused imports/variables (sdk-rpc-client-v2.ts, tui/commands/config.ts, check-import-boundaries.mjs), floating promises (sdk-rpc-client-v2.ts, fullCompaction.test.ts). - Add compactionOps.ts (model_display) to .github/FORK_OWNED_FILES so the relocated compaction-model fields are guard-listed. Verification: - pnpm run typecheck green; pnpm run lint at fork baseline (no new findings vs pre-merge HEAD). - FORK_OWNED_FILES markers all resolve. - Fork suites pass: TUI commands + session CLI (75 tests), x-opencode-session (7), update-all-session-models flag, oauth openai-compatible. - fullCompaction.test.ts: 2 failures, both pre-existing at pre-merge HEAD (baseline had 5; upstream's macOS stabilization fixed 3. EOF )
1 participant
Related Issue
N/A — internal task: client-side phase 1 support for the OpenGW "thinking hiding + summary" rollout (no public issue).
Problem
OpenGW will hide raw model thinking from Chat Completions responses and instead send a readable summary plus an encrypted reasoning state that clients must replay verbatim on later turns. The wire shape is a new
reasoning_detailsarray (summary segments plus one trailing encrypted element, index-addressed when streaming) alongside the existingreasoning_contentplaintext. Today kimi-code silently drops array-shapedreasoning_details— the reasoning extractor only accepts strings — and outbound history only echoes concatenated think text under a single reasoning key while discardingThinkPart.encrypted. Once OpenGW makesreasoning_detailsmandatory on continuation (phase 2 rejects requests without it), multi-turn conversations would fail.What changed
reasoning_detailsis parsed intoThinkParts keyed by the element'sindex; the string reasoning dialect is only observed when no array is present. Elements whosetypeis notsummary/encryptedare skipped, so OpenRouter'sreasoning.*array dialect stays ignored as before.ThinkPartgainsdetailsIndex;mergeInPlacerefuses to merge think parts with different indices, so summary segment boundaries survive streaming and the array replays in original order (encrypted element last, delivered once).reasoning_details(summary segments in order plus the trailing encrypted element) together withreasoning_content(segment concatenation); messages without details keep the existing dialect behavior byte-for-byte.ThinkPartshape change.Verification: focused kosong contract/provider and agent loop tests pass; the full agent-core-v2 suite passes except pre-existing environment-dependent region-marker tests on this machine; repo-wide typecheck and lint pass.
Checklist
/approve).gen-changesetsskill, or this PR needs no changeset.gen-docsskill, or this PR needs no doc update.