Fix stale runtime model reuse on session reset by PonyX-lab · Pull Request #41173 · openclaw/openclaw

PonyX-lab · 2026-03-09T15:18:04Z

Describe the problem and fix in 2–5 bullets:

Problem: sessions.reset reused stale runtime model / modelProvider fields from the previous session entry instead of recomputing from current defaults and explicit
overrides.
Why it matters: after changing the configured default model, resetting an existing session could keep the session pinned to an outdated or unsupported model.
What changed: gateway reset now strips stale runtime model state before calling resolveSessionModelRef(...), and runtime resetSession() also clears stale runtime
model metadata defensively.
What did NOT change (scope boundary): model resolution precedence for normal non-reset flows is unchanged; this only affects reset paths.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Closes #
Related #

User-visible / Behavior Changes

sessions.reset now recomputes the next session model from current defaults and session overrides instead of preserving stale runtime model metadata from the previous
session.

Security Impact (required)

New permissions/capabilities? (No)
Secrets/tokens handling changed? (No)
New/changed network calls? (No)
Command/tool execution surface changed? (No)
Data access scope changed? (No)
If any Yes, explain risk + mitigation:

Repro + Verification

Environment

OS: Linux
Runtime/container: local repo checkout with pnpm
Model/provider: reproduced with qwencode/qwen3.5-plus-2026-02-15 stale runtime state and openai/gpt-test-a configured default in tests
Integration/channel (if any): none
Relevant config (redacted): agent default model changed after existing session already persisted runtime model fields

Steps

Create or keep a session entry whose persisted runtime fields contain an old modelProvider / model.
Change the configured default model.
Call sessions.reset for that session.

Expected

Reset session resolves model from current defaults plus any explicit session overrides.

Actual

Reset session kept using the previous session's stale runtime model metadata.

Evidence

Attach at least one:

Failing test/log before + passing after
Trace/log snippets
Screenshot/recording
Perf numbers (if relevant)

Human Verification (required)

What you personally verified (not just CI), and how:

Verified scenarios: ran targeted gateway and e2e tests covering gateway sessions.reset and runtime resetSession() retry behavior.
Edge cases checked: explicit reset path recomputes from defaults; compaction-failure retry path clears stale runtime model fields from persisted session state.
What you did not verify: full pnpm build && pnpm check && pnpm test repo-wide suite.

Review Conversations

I replied to or resolved every bot review conversation I addressed in this PR.
I left unresolved only the conversations that still need reviewer or maintainer judgment.

If a bot review conversation is addressed by this PR, resolve that conversation yourself. Do not leave bot review conversation cleanup for maintainers.

Compatibility / Migration

Backward compatible? (Yes)
Config/env changes? (No)
Migration needed? (No)
If yes, exact upgrade steps:

Failure Recovery (if this breaks)

How to disable/revert this change quickly: revert commit f07e5a0.
Files/config to restore: src/gateway/server-methods/sessions.ts, src/auto-reply/reply/agent-runner.ts.
Known bad symptoms reviewers should watch for: reset sessions unexpectedly ignoring explicit model overrides (not expected; covered by existing model resolution
precedence and targeted tests).

Risks and Mitigations

Risk: reset paths might accidentally drop intended sticky model configuration.
Mitigation: only runtime model / modelProvider / systemPromptReport are cleared; explicit modelOverride / providerOverride remain intact, and targeted tests
cover reset recomputation behavior.

AI-assisted: yes.
Testing: targeted tests only.

greptile-apps · 2026-03-09T15:21:41Z

Greptile Summary

This PR fixes a bug where sessions.reset reused stale runtime model/modelProvider/systemPromptReport fields from the previous session entry instead of recomputing them from current defaults and explicit overrides. The fix is applied in two places: the gateway sessions.reset handler (which now strips runtime model state before calling resolveSessionModelRef), and the runtime resetSession() function in agent-runner.ts (which defensively clears stale runtime model fields on compaction-failure retries). Targeted tests are added for both paths.

src/gateway/server-methods/sessions.ts: New stripRuntimeModelState() helper strips model, modelProvider, and systemPromptReport from an entry before passing it to resolveSessionModelRef(...), ensuring the resolved model for the new session comes from current defaults/overrides rather than stale runtime metadata.
src/auto-reply/reply/agent-runner.ts: resetSession() now explicitly sets model, modelProvider, and systemPromptReport to undefined when constructing the next entry, preventing stale runtime values from being spread into the reset entry.
Tests: A gateway integration test confirms that after resetting a session with a stale qwencode model, the new entry reflects the currently configured openai/gpt-test-a default. An e2e test confirms that on compaction-failure retry, stale runtime model fields are cleared from both the in-memory store and the persisted JSON.
The fix correctly preserves modelOverride and providerOverride through the strip step, so explicit per-session model pinning continues to work as expected on reset.

Confidence Score: 5/5

This PR is safe to merge — the change is narrowly scoped to reset paths, preserves all explicit override fields, and is verified by targeted tests.
The fix is minimal and well-understood, directly addressing the reported stale-state bug. stripRuntimeModelState only touches the three runtime-only fields (model, modelProvider, systemPromptReport) and intentionally leaves modelOverride/providerOverride intact, which is exactly the right invariant. The resolveSessionModelRef code confirms it correctly falls through to those override fields when runtime fields are absent. Both code paths are covered by tests, and no normal (non-reset) flows are affected.
No files require special attention.

_{Last reviewed commit: f07e5a0}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f07e5a00c4

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".

src/gateway/server-methods/sessions.ts

jalehman · 2026-03-10T21:02:46Z

Merged via squash.

Prepared head SHA: d8a04a4
Merge commit: 5337439

Thanks @PonyX-lab!

* main: (42 commits) test: share runtime group policy fallback cases refactor: share windows command shim resolution refactor: share approval gateway client setup refactor: share telegram payload send flow refactor: share passive account lifecycle helpers refactor: share channel config schema fragments refactor: share channel config security scaffolding refactor: share onboarding secret prompt flows refactor: share scoped account config patching feat(discord): add autoArchiveDuration config option (openclaw#35065) fix(gateway): harden token fallback/reconnect behavior and docs (openclaw#42507) fix(acp): strip provider auth env for child ACP processes (openclaw#42250) fix(browser): surface 429 rate limit errors with actionable hints (openclaw#40491) fix(acp): scope cancellation and event routing by runId (openclaw#41331) docs: require codex review in contributing guide (openclaw#42503) Fix stale runtime model reuse on session reset (openclaw#41173) docs: document r: spam auto-close label fix(ci): auto-close and lock r: spam items fix(acp): implicit streamToParent for mode=run without thread (openclaw#42404) test: extract sendpayload outbound contract suite ...

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

Fixes openclaw#43930 When switching between models (e.g., Gemini → Grok → Gemini), the context may contain function_response entries that were valid for the previous model but cause errors for Gemini API with 'Name cannot be empty'. This fix skips tool calls without valid name during message conversion, preventing the INVALID_ARGUMENT error from Gemini API. Related to PR openclaw#41173 which added stripRuntimeModelState() for session reset, but model switching is a different scenario that needs separate handling.

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman (cherry picked from commit 5337439)

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman (cherry picked from commit 5337439)

@jalehman

Merged via squash. Prepared head SHA: d8a04a4 Co-authored-by: PonyX-lab <[email protected]> Co-authored-by: jalehman <[email protected]> Reviewed-by: @jalehman (cherry picked from commit 5337439)

openclaw-barnacle bot added gateway Gateway runtime size: S labels Mar 9, 2026

chatgpt-codex-connector bot reviewed Mar 9, 2026

View reviewed changes

src/gateway/server-methods/sessions.ts Show resolved Hide resolved

jalehman self-assigned this Mar 9, 2026

jalehman force-pushed the fix/sessions-reset-stale-runtime-model branch 5 times, most recently from 8f35b25 to 083726d Compare March 10, 2026 21:00

小马哥 and others added 3 commits March 10, 2026 14:01

Fix stale runtime model reuse on session reset

f7137ae

Clear stale context window on session reset

61affb9

fix: recompute session reset model state

d8a04a4

jalehman force-pushed the fix/sessions-reset-stale-runtime-model branch from 083726d to d8a04a4 Compare March 10, 2026 21:02

jalehman merged commit 5337439 into openclaw:main Mar 10, 2026
3 checks passed

github-actions bot mentioned this pull request Mar 10, 2026

📡 Upstream Digest — 2026-03-10 22:20 UTC curtismercier/openclaw-mods#230

Open

github-actions bot mentioned this pull request Mar 12, 2026

上游更新: v2026.3.11 — 14 P0 + 32 P1 待合并 jiulingyun/openclaw-cn#499

Open

guang384 mentioned this pull request Mar 12, 2026

[Bug]: LLM Error switching default agent from Gemini #43930

Open

guang384 mentioned this pull request Mar 12, 2026

fix(agents): skip tool calls without valid name on model switch #43949

Closed

guang384 mentioned this pull request Mar 12, 2026

fix(agents): skip tool calls without valid name on model switch #43985

Open

alexey-pelykh mentioned this pull request Mar 23, 2026

Cherry-pick: Auto-reply and conversation handling (1/2) (50 commits) remoteclaw/remoteclaw#1896

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix stale runtime model reuse on session reset#41173

Fix stale runtime model reuse on session reset#41173
jalehman merged 3 commits intoopenclaw:mainfrom
PonyX-lab:fix/sessions-reset-stale-runtime-model

PonyX-lab commented Mar 9, 2026

Uh oh!

greptile-apps bot commented Mar 9, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

Uh oh!

Uh oh!

jalehman commented Mar 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

PonyX-lab commented Mar 9, 2026

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Review Conversations

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Uh oh!

greptile-apps bot commented Mar 9, 2026

Greptile Summary

Confidence Score: 5/5

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

jalehman commented Mar 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants