Skip to content

fix: omit empty tools from OpenAI compaction requests - #199

Merged
hetaoBackend merged 5 commits into
mainfrom
fix/compaction-empty-tools
Oct 1, 2026
Merged

fix: omit empty tools from OpenAI compaction requests#199
hetaoBackend merged 5 commits into
mainfrom
fix/compaction-empty-tools

Conversation

@hetaoBackend

@hetaoBackend hetaoBackend commented Sep 19, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator

Change

Fixes #194. Compaction checkpoints preserve historical tool calls/results but supply no tool definitions. The OpenAI Completions adapter added tools: [] solely because of that history, causing backends such as vLLM to reject the request with HTTP 400.

Omit tools when definitions are absent or empty. Keep nonempty tool definitions unchanged. The runtime change removes the history-based empty-array injection and its helper; it adds no configuration or recovery requests.

This follows the optional tools field in the OpenAI API. Anthropic also removed the requirement to supply tools alongside historical tool blocks on February 27, 2025. The vendor ledger and existing regression fixture are updated, and a distribution-owned HTTP integration test is registered in the suite and source inventory.

Validation

  • PASS: 7 focused HTTP regressions covering compaction through custom/OpenAI provider paths, preserved tool history, absent/empty definitions, Anthropic cache compatibility, ordinary requests, and nonempty tools.
  • PASS: git diff --check and full pnpm verify on clean commit 237f0504324034dbb9be3447bc654f166967c7ae, macOS arm64 / Node 26.4.0; all 14 gates passed.
  • NOT RUN: live LiteLLM/vLLM or other provider acceptance; Linux/Windows locally; interactive automatic-threshold triggering. Admission sizing is synthetic; the tests exercise the real compaction algorithm, checkpoint provider, SDK serialization, local HTTP transport, and stream parser. Gateway-specific requirements to resupply tool definitions are outside this fix.

Publication and contribution checks

  • I have permission to contribute these changes under the existing licenses applicable to the changed files/packages; imported material and its provenance are identified and existing notices are preserved.
  • No credentials, account data, real user content, internal source history or private review material is included.
  • Added/removed source files were reviewed before regenerating release/public-source.json; new tests are declared in test/vitest-suites.json where applicable.
  • Shared English/Chinese documentation and capability/verification records are updated where applicable. Mock/offline results are not described as live-service acceptance.

Maintainer handoff

Publication scope or license changes: none beyond the declared synthetic regression test.

Shared-source port: pending. Upstream vendor PR: not created.

@hetaoBackend hetaoBackend mentioned this pull request Sep 19, 2026
Closed
@hetaoBackend hetaoBackend changed the title fix: omit empty tools on OpenAI compaction requests fix: negotiate empty tools for OpenAI compaction requests Sep 19, 2026
@hetaoBackend hetaoBackend changed the title fix: negotiate empty tools for OpenAI compaction requests fix: omit empty tools from OpenAI compaction requests Sep 19, 2026
yujiachen-y
yujiachen-y previously approved these changes Sep 19, 2026
View reviewed changes

@yujiachen-y yujiachen-y left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM 🚀

@SaladDay

SaladDay commented Sep 19, 2026

Copy link
Copy Markdown
Collaborator

LGTM

@hetaoBackend hetaoBackend added bug Something isn't working cli Standalone mcode: TUI, headless, ACP and source builds/tooling labels Sep 23, 2026
@hetaoBackend
hetaoBackend merged commit bf3916f into main Oct 1, 2026
13 checks passed
@hetaoBackend
hetaoBackend deleted the fix/compaction-empty-tools branch October 1, 2026 02:16
@tournierjc tournierjc mentioned this pull request Oct 1, 2026
Merged
4 tasks done
tournierjc added a commit to tournierjc/kinetick-code that referenced this pull request Oct 1, 2026
@tournierjc @SaladDay @hetaoBackend
merge: upstream c08148e (TPS timing, empty OpenAI tools, stable regul…
7b5dee8 
…ar-mode history) (#108)

* fix: include hidden thinking in TPS timing (MiniMax-AI#413)

* fix: include hidden thinking in TPS timing

* docs: note thinking_start in first token timing comment

* test(windows): allow the NTFS source check to exceed the default timeout

---------

Co-authored-by: hetaoBackend <hetao7@pku.edu.cn>

* fix: omit empty tools from OpenAI compaction requests (MiniMax-AI#199)

* fix: omit empty tools on OpenAI compaction requests

* fix: negotiate missing tools with bounded compatibility recovery

* fix: simplify empty tools handling to omit absent definitions

* fix(tui): keep regular-mode history stable during live runs (MiniMax-AI#416)

Regular mode wrote rows into native scrollback as soon as they scrolled
off screen. When such a row later changed (a parallel tool finishing, a
streamed table widening, a list turning loose, a turn growing past the
projection fold, the welcome status badge flipping, or a stopped prompt
gaining its cancelled marker), the renderer cleared scrollback and
replayed the whole session from the welcome logo.

Only final rows now reach native history:

- The transcript reports how many leading rows later updates cannot
  change, including the committed blocks of a streaming Markdown reply.
- The chat layout keeps rows that may still change within one screen
  above the footer, showing their latest rows under a one-line notice
  when they do not fit.
- The regular-mode projection only appends; established turns are not
  re-folded or dropped.
- The welcome banner above a conversation is static; account notices
  that need action appear above the Composer.
- Stopping a run no longer marks its delivered prompt as cancelled; the
  marker stays reserved for prompts returned to the Composer.

To keep long sessions responsive, the transcript drops final rows far
above the screen and the engine rebases its retained state
(takeDiscardedRows) instead of rewriting history.

Co-authored-by: tournierjc <tournierjc@users.noreply.github.com>

---------

Co-authored-by: SaladDay <1203511142@qq.com>
Co-authored-by: hetaoBackend <hetao7@pku.edu.cn>
Co-authored-by: DanielWalnut <45447813+hetaoBackend@users.noreply.github.com>
Co-authored-by: tournierjc <tournierjc@users.noreply.github.com>
PAKIKNOWLEDGE added a commit to PAKIKNOWLEDGE/Talos-Code-CLI that referenced this pull request Oct 1, 2026
@PAKIKNOWLEDGE @hetaoBackend
fix: port approved upstream transport and transcript fixes
953cb94 
Source: MiniMax-AI/minimax-code PR MiniMax-AI#199 (bf3916f), MiniMax-AI#388 (67a5c2a), and literal-text only from MiniMax-AI#410 (5683465).

Co-authored-by: DanielWalnut <45447813+hetaoBackend@users.noreply.github.com>
@tournierjc tournierjc mentioned this pull request Oct 1, 2026
Merged
tournierjc added a commit to tournierjc/kinetick-code that referenced this pull request Oct 3, 2026
@tournierjc @AdhereZ
minimax @hetaoBackend @amszuidas chenhao @SaladDay @hermes-agent @cursoragent
merge: upstream 564e916 (MiniMax-AI#423 side-session Esc and run time…
b084d3a 
…r, on 0.6.0/0.6.1 sync) (#110)

* fix: remove forced branching from bash guidance (MiniMax-AI#358)

Co-authored-by: minimax <adhere@minimaxi.com>

* feat: sync reviewed 0.5.5 runtime improvements (MiniMax-AI#367)

Preserve public privacy defaults and distribution boundaries while porting compaction, Bash timeout, memory output and update-proxy improvements. Record selective provenance without advancing the full source baseline.

Assisted-by: codex-cli reason:public-source-sync-0.5.5

* docs: publish feedback and private security contact channels (MiniMax-AI#360)

* docs: publish feedback and private security contact channels

* docs: correct MiniMax Agent X contact link

* fix(tui): distinguish Bash recaps and import models during onboarding (MiniMax-AI#369)

* fix: measure token speed over model generation time (MiniMax-AI#373)

* chore: bump version to 0.5.6 (MiniMax-AI#374)

* feat(config): add M3.1 Flash Preview to the fallback catalog (MiniMax-AI#375)

Include context window and effort controls in the first-run catalog while preserving existing managed snapshots.

Docs-Impact: Existing model selection workflow is unchanged.

Assisted-by: codex-cli reason:m31-flash-fallback-sync

* chore: bump version to 0.5.7 (MiniMax-AI#377)

* chore: bump version to 0.5.7

* test: drain stderr before nonzero Bash fixture exit

* fix(tui): display hook system messages without adding model context (MiniMax-AI#376)

* fix(tui): anchor restored sessions above inherited terminal history (MiniMax-AI#378)

* fix(tui): keep slow terminal output off the input loop (MiniMax-AI#379)

* chore: bump version to 0.5.8 (MiniMax-AI#380)

* feat: cascade explicit session stops to owned background work (MiniMax-AI#381)

Port the reviewed shared runtime behavior while retaining standalone composition. Keep conversation leave separate from explicit stop, suppress canceled task delivery, and track append activations through teardown.


Assisted-by: codex-cli reason:selective-runtime-port
Docs-Impact: document stop and session-leave behavior in docs/tui-capabilities.md

Co-authored-by: chenhao <chenhao@minimaxi.com>

* fix: query structured Windows source volume properties (MiniMax-AI#383)

* feat(examples): add Pocket Pet demo and refresh README onboarding (MiniMax-AI#385)

* feat(examples): add reproducible Pocket Pet demo and refresh onboarding

* feat(examples): polish Pocket Pet interaction and demo

* fix(tui): dismiss stale terminal errors when a new turn starts (MiniMax-AI#388)

* fix(tui): preserve native scrollback during automatic updates (MiniMax-AI#389)

* fix(tui): preserve native scrollback during projection updates

* fix(tui): isolate transient overlays from native scrollback

* fix(tui): stabilize native history and diff overlay frames (MiniMax-AI#390)

* fix(tui): prevent idle transcript replay and diff overlay frames

* fix(tui): retain prompt identity and deferred main-buffer geometry

* test: pin the srt-macos probe PATH and scale the BYOK serial timeout budget (MiniMax-AI#397)

* test(sandbox): pin the probe PATH instead of inheriting the ambient one

The real sandbox-exec probes resolve their command through PATH, so
inheriting the caller's let any wrapper ahead of /bin decide what `rm`
means. On a machine that has run the agent, that is the recoverable-delete
shim MCode installs into agent shells: `rm` resolves to the shim, the shim
execs mavis-trash, the sandbox denies that trash move as a write outside
the policy under test, and a correct allow-probe exits 1. The suite then
reports a sandbox regression where the sandbox behaved correctly.

Pin all three probe environments to the system binaries the probes
actually use, matching the wrapper-argument test above them. This also
drops the unguarded `process.env.PATH` in wrapProfile, which had no
fallback and could hand a real sandbox spawn an undefined PATH.

Fixes MiniMax-AI#394.

* test(byok): scale the serial BYOK budget by platform

The first case is 30+ serial CLI spawns, each booting the whole runtime,
so its wall-clock cost tracks machine speed rather than the transport it
asserts. The flat 90s budget only held with roughly 1.5x headroom on an
idle machine, so a parallel build, a laptop under load, or a shared CI
runner turned a passing transport into a `testTimeoutFailure` that named a
product failure no assertion had actually observed.

Budget it the way smoke.test.mjs already budgets runtime startup: a base
allowance plus a Windows multiplier for slower process spawn. The test is
not slowed down by a larger ceiling; it only stops failing for reasons
unrelated to what it checks.

Fixes MiniMax-AI#395.

* feat: sync MiniMax Code 0.5.9 runtime and TUI changes (MiniMax-AI#398)

TUI
- Report active root-session background tasks as `background=N` in the
  `[V]` build-mode status line. Bash still owned by its foreground tool
  call is excluded, and the count holds at 1 after a turn settles until a
  fresh task list arrives. The documented minimum width is now 102 columns.

Runtime
- The MiniMax API-key route now shares the official model catalog with the
  managed route; the first-run catalog moves to `minimax-model-catalog.ts`
  with unchanged model definitions.
- Apply byte limits to large media and accumulated history for BYOK requests.
- Send the M3 thinking toggle as a thinking setting rather than a generic
  reasoning effort.
- Allow edit and rewind after an interrupt that left only a background
  reminder in canonical history.
- Read less data when listing session files and navigating long histories.
- Measure token output rate from the time events are observed.
- Add `worktreeRefreshBeforeCreate` (default on) to fetch the selected
  upstream before creating a worktree, and exclude nested roots from fork
  worktree fingerprints.
- Add Ghostty to the external editor catalog.
- Enable the Codex OAuth model settings entry by default.

Prompts
- Cron guidance now lives only in tool definitions; the legacy feature
  template is empty. Memory edits use the `memory` tool's `edit` operation.
- Clarify multimodal tool discovery in the mcode-tools reminder.

* chore: release MiniMax Code 0.5.9 (MiniMax-AI#399)

* Revert MiniMax-AI#390 and MiniMax-AI#389 to restore TUI scrollback baseline (MiniMax-AI#396)

* Revert "fix(tui): stabilize native history and diff overlay frames (MiniMax-AI#390)"

This reverts commit 1e136bf.

* Revert "fix(tui): preserve native scrollback during automatic updates (MiniMax-AI#389)"

This reverts commit 9b9c07b.

* fix(tui): sync 0.5.10 prompt and run recovery fixes (MiniMax-AI#410)

* chore: release MiniMax Code 0.5.10 (MiniMax-AI#411)

* fix: include hidden thinking in TPS timing (MiniMax-AI#413)

* fix: include hidden thinking in TPS timing

* docs: note thinking_start in first token timing comment

* test(windows): allow the NTFS source check to exceed the default timeout

---------

Co-authored-by: hetaoBackend <hetao7@pku.edu.cn>

* fix: omit empty tools from OpenAI compaction requests (MiniMax-AI#199)

* fix: omit empty tools on OpenAI compaction requests

* fix: negotiate missing tools with bounded compatibility recovery

* fix: simplify empty tools handling to omit absent definitions

* fix(tui): keep regular-mode history stable during live runs (MiniMax-AI#416)

Regular mode wrote rows into native scrollback as soon as they scrolled
off screen. When such a row later changed (a parallel tool finishing, a
streamed table widening, a list turning loose, a turn growing past the
projection fold, the welcome status badge flipping, or a stopped prompt
gaining its cancelled marker), the renderer cleared scrollback and
replayed the whole session from the welcome logo.

Only final rows now reach native history:

- The transcript reports how many leading rows later updates cannot
  change, including the committed blocks of a streaming Markdown reply.
- The chat layout keeps rows that may still change within one screen
  above the footer, showing their latest rows under a one-line notice
  when they do not fit.
- The regular-mode projection only appends; established turns are not
  re-folded or dropped.
- The welcome banner above a conversation is static; account notices
  that need action appear above the Composer.
- Stopping a run no longer marks its delivered prompt as cancelled; the
  marker stays reserved for prompts returned to the Composer.

To keep long sessions responsive, the transcript drops final rows far
above the screen and the engine rebases its retained state
(takeDiscardedRows) instead of rewriting history.

* feat: sync and release MiniMax Code 0.6.0 (MiniMax-AI#421)

Port the reviewed 0.6.0 runtime and TUI behavior into the standalone
distribution and bump the release version to 0.6.0.

- Allow /retry inside a /btw side conversation to resend its last message.
- Default to MiniMax-M3.1-Flash-Preview when no model has been selected.
- After an accepted Goal completion, keep the Turn open for one final reply
  that summarizes the result and deliverables; blocked proposals still end
  the Turn.
- Do not retry provider safety refusals, including on BYOK, and classify
  them as content_filter; keep Anthropic refusal details in the error.
- Retry TLS record verification failures before output like other
  transient network errors.
- Keep a post-compaction reminder at the tail of a continuation instead of
  aborting the provider call and recompacting on every retry.

* feat: sync and release MiniMax Code 0.6.1 (MiniMax-AI#422)

Allow /doctor and /feedback in a /btw side conversation so a failed side
response can be diagnosed or reported without leaving it. Neither command
mutates the parent or side Session. /quit stays blocked in the side view,
because leaving from there aborts only the side Turn and skips side
Session cleanup.

Bump the root and TUI source versions to 0.6.1.

* test(tui): adopt upstream MiniMax-AI#410 literal-prompt assertion in compact-rail test

Upstream 5683465 (MiniMax-AI#410) changed the 'renders intent and execution as one
compact visual rail' expectation to keep the prompt's ** markers literal
(fork prompt-literal behavior). The three-way merge kept the pre-MiniMax-AI#410
expectation; adopt the MiniMax-AI#410 form.

* fix(config): adopt upstream MiniMax-AI#421 default model MiniMax-M3.1-Flash-Preview

Upstream 9150441 (sync 0.6.0) flipped buildPresetEntry's defaultModel to
minimax/MiniMax-M3.1-Flash-Preview; the three-way merge kept the main side
for config.ts and dropped the flip, breaking the MiniMax-AI#421-pinned
model-selection/management suites. Restore the upstream value.

* fix(tui): keep side-session Esc and main run timer consistent (MiniMax-AI#423)

In a /btw side conversation, an empty Escape no longer arms the
double-Escape /edit shortcut or shows its "Press Esc again to edit"
hint, since /edit is unavailable in side conversations. Escape still
interrupts a live side response.

Switching between the side and main views re-adopted the main Session's
live Turn at the switch time, which reset the Running/Loading timer and
shortened the settled Turn duration. A Turn start ledger now records the
earliest observed start (submission, session.start events including
hidden Sessions, and adoption), and both re-adoption and the activity
line use it.

Bump the root and TUI source versions to 0.6.2.

* fix(tui): adopt upstream MiniMax-AI#423 side-session Esc and turn start ledger

Record the earliest Turn start across side-session switches so the main
run timer and settled duration stay stable, and ignore double-Escape
/edit while a side conversation is open. Fork packages stay kinetick-code
and @mavis/code at 0.6.9, and multi-tab live turn watchers stay in place.

Co-authored-by: tournierjc <tournierjc@users.noreply.github.com>

* test(tui): drop duplicate applyPendingModelSelection key in command-flow mock

The duplicate key dates from the #55/#57 merge on origin/main (5bba9c7): both
sides added the same mock key. vitest transforms tolerate it; ESLint does not.

---------

Co-authored-by: AdhereZ <85055734+AdhereZ@users.noreply.github.com>
Co-authored-by: minimax <adhere@minimaxi.com>
Co-authored-by: DanielWalnut <45447813+hetaoBackend@users.noreply.github.com>
Co-authored-by: AmsZuidas <254873068+amszuidas@users.noreply.github.com>
Co-authored-by: chenhao <chenhao@minimaxi.com>
Co-authored-by: SaladDay <1203511142@qq.com>
Co-authored-by: hetaoBackend <hetao7@pku.edu.cn>
Co-authored-by: hermes-agent <hermes-agent@users.noreply.github.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: tournierjc <tournierjc@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@yujiachen-y yujiachen-y yujiachen-y left review comments

Assignees

No one assigned

Labels

bug Something isn't working cli Standalone mcode: TUI, headless, ACP and source builds/tooling

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

tools: [] sent on automatic-compaction requests → HTTP 400 from OpenAI-compatible backends (vLLM); compaction never succeeds

3 participants

@hetaoBackend @SaladDay @yujiachen-y