OpenClaw merged a small but useful agent-runtime fix at the edge of the August 29th morning scan window. PR #131444, merged at 07:58 UTC, preserves the Gateway's valid queued run status when agents call the sessions_list tool.
The issue was a projection mismatch. The Gateway already knew when an admitted run was waiting for a concurrency slot, and its session row could represent that state as queued. But the model-facing sessions_list output dropped that value, so the agent saw no status for a child session that was actually admitted and waiting.
That made queued work look too much like idle work.
The Gateway Schema Becomes The Source Of Truth
The repair uses the Gateway protocol's SessionRunStatusSchema as the shared runtime owner across the relevant boundaries: the Gateway row, the model-facing output schema, and the agent-tool narrowing check. Instead of maintaining separate allowlists that can drift, sessions_list now follows the same status vocabulary the Gateway already emits.
The change is intentionally narrow. The PR says there are no configuration, environment-variable, storage, migration, dependency, compatibility, scheduling, or security changes. Production scope is actually smaller after the repair, because duplicate status handling was removed.
Why Queued Status Matters
OpenClaw agents increasingly coordinate child sessions and background work. When a child run is admitted but waiting behind a concurrency limit, that is a real state the parent agent needs to understand.
Without queued, an agent might assume the child session has no active run and dispatch duplicate work, report the wrong status, or tell a user that nothing is happening. With the status preserved, agents can make a cleaner distinction between waiting, running, completed, and idle work.
The practical impact is simple:
- Queued child sessions now show
status: "queued". - Agents can avoid duplicate dispatch while concurrency slots are full.
- User-facing progress reports can describe waiting work more accurately.
- The Gateway and agent-tool schemas stay aligned.
Tested Against A Real Gateway Path
The proof used a real Gateway process, the real agent CLI, two agent turns, and a mock OpenAI-compatible provider. With agents.defaults.maxConcurrent=1, the first turn occupied the only slot while the second was admitted and queued. The first model response then invoked the real sessions_list tool.
Before the fix, the Gateway row showed queued, but the tool output showed the status as missing. After the fix, both showed queued.
Focused validation covered the Gateway protocol test suite, the sessions-list tool tests, formatting, Oxlint, changed-scope repository guards, exact-main and exact-head builds, and git diff --check.
For operators, this is not a flashy feature, but it removes ambiguity from multi-agent orchestration. OpenClaw already had the right state; now the agent-facing tool preserves it all the way to the model.