fix(workboard): record resolved runtime metadata instead of hardcoded codex engine#108887
Conversation
… codex engine
Workboard executions labeled every dispatched run engine=codex, model=default,
and id suffix :codex even for Claude/other harness agents. The gateway agent
admission phase now returns the resolved {harness, provider, model} for plugin
subagent runs; the dispatcher records it verbatim and omits engine/model when
unresolved. Engine becomes an open runtime identifier in the workboard
contract (built-in launch choices stay a closed list), store/UI normalizers
preserve historical labels as written instead of inventing codex, and new
execution ids use an :agent-session suffix. Fixes #108362
openclaw-barnacle
Bot
added
app: web-ui
![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: 0408d71cc7
ℹ️ About Codex in GitHub
Your team has set up Codex to 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 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| provider: effectiveProviderOverride ?? sessionModel.provider, | ||
| model: sessionModel.model, | ||
| }; |
There was a problem hiding this comment.
For plugin subagent runs without an explicit model override, this reports the session/default provider and model before the command's model-selection pipeline runs. I checked src/agents/command/model-selection.ts:235-269: channels.modelByChannel can still replace the provider/model from the session entry or run context, and src/agents/command/run-embedded-attempt.ts:308-311 observes the actual attempt later. In a channel-backed subagent session with a channel model override, Workboard will persist the default runtime while the run executes the channel-selected model, recreating false audit metadata.
Useful? React with 👍 / 👎.
| ...(params.resolvedSessionKey === "global" ? { agentId: params.activeSessionAgentId } : {}), | ||
| status: "accepted" as const, | ||
| acceptedAt: Date.now(), | ||
| ...(taskTrackingMode === "plugin_subagent" ? { runtime: resolvedRuntime } : {}), |
There was a problem hiding this comment.
This adds runtime to the accepted payload, but idempotent repeats of the same plugin subagent start hit the preflight dedupe path in src/gateway/server-methods/agent-request-preflight.ts:166-190, which reconstructs an in_flight response with only run/session/agent fields and drops cached.payload.runtime. A plugin retrying a start after the first accepted response will therefore get { runId } with no runtime, so callers like Workboard omit the engine/model even though the original accepted payload had it.
Useful? React with 👍 / 👎.
|
Merged via squash.
|
… codex engine (openclaw#108887) * fix(workboard): record resolved runtime metadata instead of hardcoded codex engine Workboard executions labeled every dispatched run engine=codex, model=default, and id suffix :codex even for Claude/other harness agents. The gateway agent admission phase now returns the resolved {harness, provider, model} for plugin subagent runs; the dispatcher records it verbatim and omits engine/model when unresolved. Engine becomes an open runtime identifier in the workboard contract (built-in launch choices stay a closed list), store/UI normalizers preserve historical labels as written instead of inventing codex, and new execution ids use an :agent-session suffix. Fixes openclaw#108362 * fix(workboard): accept undefined engine in ui engineModel helper
1 participant
What Problem This Solves
Workboard recorded every dispatched agent session as
engine: codex,model: default, with an execution id ending:codex— even when the assigned agent actually ran Anthropic Claude via claude-cli. Operators auditing workboard cards saw Codex executions that never happened. (#108362)Why This Change Was Made
Execution metadata was fabricated at dispatch time before runtime resolution. The fix carries the prepared fact forward from the owning seam instead of re-deriving or guessing in the plugin:
{harness, provider, model}for plugin-subagent runs (same resolution the run itself uses: session override → model policy → implicit routing) and returns it on the run-start payload; the gateway subagent runtime exposes it as an optional, normalizedruntimeonSubagentRunResult(additive Plugin SDK surface; surface gate green).engine/modelentirely when unresolved — unknown is never labeledcodex. New execution ids use an:agent-sessionsuffix instead of encoding an engine name; existing persisted ids are untouched.WorkboardExecutionEnginebecomes an open runtime identifier in the contract (WORKBOARD_EXECUTION_ENGINESremains only the built-in launch-choice list); store and UI normalizers preserve historical labels as written — old rows minted by the bug cannot be safely reinterpreted, so they are neither rewritten nor re-defaulted (decision recorded in a code comment).User Impact
Workboard cards now show the engine/model that actually ran (e.g.
claude-cli+anthropic/claude-opus-4-8), or no engine claim at all — never a falsecodex/defaultlabel.Evidence
pnpm plugin-sdk:surface:checkgreen (additive optional field only).Fixes #108362