feat: steer mid-turn prompts by default by fuller-stack-dev · Pull Request #77023 · openclaw/openclaw

fuller-stack-dev · 2026-05-04T01:19:38Z

Summary

Make steer the clear default active-run queue mode: normal inbound messages try to enter the active run instead of always waiting for a later turn.
Preserve the explicit queue-by-default modes people can choose: followup queues each message for later, collect coalesces compatible queued messages for one later turn, and interrupt aborts the active run for the newest message.
Harden explicit /steer <message> so user text is never swallowed: if same-turn steering is unavailable, rejected, compacting, idle, or throws, OpenClaw strips /steer and continues with <message> as a normal prompt.
Remove overlapping legacy queue mode names from the public config surface while migrating them safely: queue -> steer, steer-backlog / steer+backlog -> followup.
Update schema/help/generated config docs, queue docs, slash-command docs, changelog, and targeted queue/steer regression coverage.

Before

Active-run queue behavior had overlapping modes: steer, legacy queue, followup, collect, steer-backlog, and interrupt.
queue meant serialized steering, while steer meant batched steering; the distinction was implementation-shaped and hard to explain.
steer-backlog both steered and preserved the same message for a later followup, which could look like duplicate handling on streaming surfaces.
/steer <message> depended on same-turn steering succeeding; some fallback paths could leave the command body in an unsafe/unclear state instead of reliably treating <message> as a normal prompt.
The docs and PR text over-framed followup/collect as fallback behavior, which made it unclear whether users could still queue by default.

After

Supported queue modes are intentionally small and user-facing:
- steer: default. Try to inject normal inbound messages into the active run; if the runtime cannot accept steering, wait for the active run to finish and run the prompt later.
- followup: do not steer; queue each message for a later turn after the active run ends.
- collect: do not steer; queue compatible messages and run one coalesced later turn after the quiet window.
- interrupt: abort the active run and start the newest message.
/queue steer controls future normal inbound messages. /steer <message> is an explicit one-off command and works independently of the stored /queue mode.
/steer <message> never drops the message text: when steering cannot happen, <message> continues as a normal prompt.
Existing configs using retired names keep working through doctor/config migration.

Verification

pnpm test src/auto-reply/reply/queue/settings.test.ts src/auto-reply/reply.directive.parse.test.ts src/auto-reply/reply/directive-handling.queue-validation.test.ts src/auto-reply/reply/get-reply-run.media-only.test.ts src/auto-reply/reply/agent-runner.media-paths.test.ts src/commands/doctor/shared/legacy-config-migrate.test.ts src/config/config-misc.test.ts src/config/schema.help.quality.test.ts src/config/config.legacy-config-detection.accepts-imessage-dmpolicy.test.ts src/auto-reply/reply/commands-steer.test.ts src/auto-reply/reply/get-reply-inline-actions.skip-when-config-empty.test.ts src/auto-reply/reply/get-reply.before-agent-reply.test.ts
pnpm config:schema:check
pnpm config:docs:check
pnpm format:docs:check
git diff --check
Testbox pnpm check:changed: run 25785282586 / tbx_01krg496nw9zq7axbjvywtcfsh, passed.

clawsweeper · 2026-05-04T01:22:56Z

Codex review: needs real behavior proof before merge.

Summary
The PR makes normal mid-run prompts attempt same-turn steering by default, narrows /queue to fallback modes, updates Codex steering batching and /steer fallback handling, and refreshes related docs/tests/changelog.

Reproducibility: not applicable. this is a feature/config-surface PR, not a bug report. Source comparison shows current main still has the old steering-mode surface while the PR head implements the new default behavior.

Real behavior proof
Needs real behavior proof before merge: The PR body claims a live gateway smoke, but no inspectable terminal output, logs, screenshot, recording, or linked artifact is attached; the contributor should add redacted proof to the PR body and request re-review if it does not trigger automatically. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, ask a maintainer to comment @clawsweeper re-review.

Next step before merge
Human/contributor action is needed for redacted real behavior proof and maintainer acceptance of the default-steering product/config change; automation cannot supply the contributor's live setup proof.

Security
Cleared: The diff changes queue runtime behavior, config/schema/docs, doctor migrations, tests, and changelog without adding dependencies, CI execution, secrets handling, or package-resolution changes.

Review findings

[P2] Update the Codex harness queue docs — src/config/zod-schema.core.ts:455-457

Review details

Best possible solution:

Land the central queue/runtime/schema migration only after the stale Codex harness docs are corrected, maintainers accept the default-steering behavior, and redacted real Telegram or gateway proof plus appropriate broad validation is attached.

Do we have a high-confidence way to reproduce the issue?

Not applicable: this is a feature/config-surface PR, not a bug report. Source comparison shows current main still has the old steering-mode surface while the PR head implements the new default behavior.

Is this the best way to solve the issue?

Mostly yes, but not merge-ready: the PR changes the central queue resolver, runtime steering path, schema, doctor migration, and tests rather than a channel-specific workaround. It still needs the stale Codex harness docs fixed, real behavior proof, broad validation, and maintainer product approval.

Full review comments:

[P2] Update the Codex harness queue docs — src/config/zod-schema.core.ts:455-457
After this PR removes steer and queue from the public queue schema, docs/plugins/codex-harness-runtime.md still says the default is messages.queue.mode: "steer" and that legacy queue sends separate turn/steer requests. That page would publish invalid config guidance after merge, so update it with the same automatic-steering/fallback wording as the queue docs.
Confidence: 0.9

Overall correctness: patch is incorrect
Overall confidence: 0.88

Acceptance criteria:

pnpm test src/auto-reply/reply/get-reply-run.media-only.test.ts src/auto-reply/reply/agent-runner.media-paths.test.ts src/auto-reply/reply/queue/settings.test.ts src/auto-reply/reply/queue-policy.test.ts src/agents/subagent-announce-delivery.test.ts src/agents/subagent-announce.test.ts extensions/codex/src/app-server/run-attempt.test.ts src/agents/pi-embedded-runner/runs.test.ts src/auto-reply/reply.directive.parse.test.ts src/auto-reply/reply/directive-handling.queue-validation.test.ts
pnpm config:schema:check
pnpm config:docs:check
pnpm format:docs:check
git diff --check upstream/main...HEAD

What I checked:

Current main still defaults queue mode to steer: Current main returns "steer" from the queue default resolver, so the PR is not already implemented on main. (src/auto-reply/reply/queue/settings.ts:8, 40e8782400f6)
Current main still exposes removed steering modes: Current main's queue type still includes steer, queue, and steer-backlog. (src/auto-reply/reply/queue/types.ts:13, 40e8782400f6)
PR head implements default steering before fallback: The PR head sets shouldSteer for non-heartbeat, non-reset active runs whenever fallback mode is not interrupt. (src/auto-reply/reply/get-reply-run.ts:911, d62b6b1e3ac9)
PR head waits for steering acceptance: The PR head awaits queueEmbeddedPiMessageWithOutcomeAsync; accepted steering returns, while rejected steering proceeds to fallback handling. (src/auto-reply/reply/agent-runner.ts:1142, d62b6b1e3ac9)
PR head narrows the public schema: The PR head's queue schema only accepts followup, collect, and interrupt. (src/config/zod-schema.core.ts:454, d62b6b1e3ac9)
Codex harness docs remain stale on PR head: The PR head still says the Codex harness default is messages.queue.mode: "steer" and documents legacy queue, which conflicts with the schema removal. Public docs: docs/plugins/codex-harness-runtime.md. (docs/plugins/codex-harness-runtime.md:152, d62b6b1e3ac9)

Likely related people:

steipete: Current-main history ties the default queue mode, batched active steering, and active-run queue failure handling to this person across the central files. (role: feature/history owner and recent queue maintainer; confidence: high; commits: 4a6e10ece842, 30a2b3049ae0, 1ed50b0ced62; files: src/auto-reply/reply/queue/settings.ts, src/agents/pi-embedded-runner/runs.ts, src/auto-reply/reply/agent-runner.ts)
vincentkoc: Recent current-main work touched embedded-runner import/lifecycle behavior near the active-run paths affected by this PR. (role: adjacent active-run contributor; confidence: medium; commits: 4a30ae182b6a; files: src/agents/pi-embedded-runner/runs.ts)
kevinslin: Recent current-main Codex harness documentation reorganization makes this person a plausible reviewer for the stale Codex harness runtime page. (role: adjacent Codex docs contributor; confidence: medium; commits: ce0584af89bb; files: docs/plugins/codex-harness-runtime.md)

Remaining risk / open question:

The PR still lacks inspectable after-fix proof from a real gateway or Telegram setup.
The PR body says changed lanes expanded to broad/all and no Testbox broad gate was run.
The default-steering behavior and public queue-mode retirement are product/config-surface changes that need maintainer acceptance.
One Codex harness docs page still documents removed queue modes.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 40e8782400f6.

akrimm702 · 2026-05-13T10:28:48Z

Retracting this finding.

Thanks for clarifying in #77023 (comment). I treated active subagent announcements like normal inbound user prompts, but I understand now that these are internal handoffs back into the active requester turn and are intentionally kept on the steering path regardless of followup / collect queue mode.

Given that invariant, my previous P1 is not a blocker. The separate steering acceptance/rejection concern is covered by 9ed10de19c, where announce steering now awaits the async steering outcome before reporting delivery.

steipete · 2026-05-13T12:27:37Z

Thanks for rechecking. We are intentionally keeping active subagent announcements on the steering path regardless of followup / collect queue mode.

Those queue modes are user-prompt semantics. A subagent announcement is an internal handoff back into the active requester turn, so skipping steering would reintroduce the alternate delivery/queue behavior this PR is meant to remove. I added an inline note at src/agents/subagent-announce-delivery.ts in cdb36ae to make that distinction explicit.

The acceptance/rejection issue from the later review is handled separately in 9ed10de: announce steering now awaits the async steering outcome, so runtime rejection is no longer reported as delivered.

steipete · 2026-05-13T12:59:52Z

Maintainer verification for rebased head e66c9620c2397fe32dcb9f3197564ab98cf5dd62:

Rebased on current origin/main (1c5c72ea24c94140d4578fa1e2d90cbcd9be6cfd) and resolved the CHANGELOG.md entry with contributor credit.
git diff --check
rg -n "^(<<<<<<<|=======|>>>>>>>|\|\|\|\|\|\|\|)" CHANGELOG.md docs src extensions || true
pnpm test src/agents/subagent-announce-dispatch.test.ts src/agents/subagent-announce-delivery.test.ts src/agents/pi-embedded-runner/runs.test.ts src/agents/subagent-announce.format.e2e.test.ts src/agents/subagent-announce.test.ts
- passed: agents shard 71 tests, e2e shard 79 tests
pnpm test src/auto-reply/reply/commands-steer.test.ts src/auto-reply/reply/queue/settings.test.ts src/auto-reply/reply/queue-policy.test.ts src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts src/auto-reply/reply/get-reply-run.media-only.test.ts extensions/codex/src/app-server/run-attempt.test.ts -- -t "queued steering|explicit all-mode steering|flushes pending default queued steering|rejects queued steering|resolveActiveRunQueueAction|resolveQueueSettings|handleSteerCommand"
- passed: unit-fast 12 tests, auto-reply filtered 8 tests, Codex app-server filtered 6 tests

GitHub checks on the rebased head:

auto-response: passed
dispatch: passed
dependency-change-awareness: passed
label: passed
label-issues: passed
Real behavior proof: failed only because this external PR body lacks a real-behavior-proof section; local maintainer verification above covers the queue/Codex/subagent behavior touched here, so I’m treating that bot as non-blocking per maintainer policy.

Re-review progress:

State: Failed
Detail: The targeted re-review did not finish cleanly. Check the workflow run for details.
Run: https://github.com/openclaw/clawsweeper/actions/runs/25800631038
Updated: 2026-05-13T13:01:02.413Z

Summary: - Default active-run queueing to steer while preserving explicit followup/collect modes. - Keep `/steer` fallback behavior and migrate retired queue steering config. - Await Codex app-server steering acceptance so rejected/aborted steering can fall back safely. - Route active subagent announcements through intentional acceptance-aware steering, with legacy queue helpers deprecated for delivery decisions. Verification: - git diff --check - rg -n "^(<<<<<<<|=======|>>>>>>>|\|\|\|\|\|\|\|)" CHANGELOG.md docs src extensions || true - pnpm test src/agents/subagent-announce-dispatch.test.ts src/agents/subagent-announce-delivery.test.ts src/agents/pi-embedded-runner/runs.test.ts src/agents/subagent-announce.format.e2e.test.ts src/agents/subagent-announce.test.ts - pnpm test src/auto-reply/reply/commands-steer.test.ts src/auto-reply/reply/queue/settings.test.ts src/auto-reply/reply/queue-policy.test.ts src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts src/auto-reply/reply/get-reply-run.media-only.test.ts extensions/codex/src/app-server/run-attempt.test.ts -- -t "queued steering|explicit all-mode steering|flushes pending default queued steering|rejects queued steering|resolveActiveRunQueueAction|resolveQueueSettings|handleSteerCommand" Co-authored-by: fuller-stack-dev <263060202+fuller-stack-dev@users.noreply.github.com>

openclaw-barnacle Bot added docs Improvements or additions to documentation gateway Gateway runtime agents Agent runtime and tooling extensions: codex size: L labels May 4, 2026

openclaw-barnacle Bot added the commands Command implementations label May 4, 2026

This was referenced May 4, 2026

🦞 OpenClaw 生态日报 2026-05-04 borq168/radar-forge#130

Open

🦞 OpenClaw Ecosystem Digest 2026-05-04 borq168/radar-forge#133

Open

openclaw-barnacle Bot added size: XL triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. and removed size: L labels May 13, 2026

clawsweeper Bot added the mantis: telegram-visible-proof Mantis should capture Telegram visible proof. label May 13, 2026

steipete force-pushed the codex/default-midturn-steering branch from ef463c6 to 1495bfb Compare May 13, 2026 06:42

openclaw-barnacle Bot added size: L size: XL and removed size: XL size: L labels May 13, 2026

steipete self-requested a review May 13, 2026 09:47

fuller-stack-dev and others added 9 commits May 13, 2026 13:56

feat: steer mid-turn prompts by default

fd9bf0d

fix: migrate retired queue steering modes

e92215d

fix: preserve heartbeat drop before steering

b84bd19

fix: wait for steering acceptance before fallback

5fa1e83

fix: preserve steer fallback messages

45c5fac

docs: remove stale queue steer mention

6398575

fix: preserve explicit queue modes

31ebb13

fix: allow steer session queue mode

6afceb2

test: fix queue steering channel table

e76ebf5

docs: clarify steering queue deprecations

e66c962

steipete force-pushed the codex/default-midturn-steering branch from 6eb49b8 to e66c962 Compare May 13, 2026 12:58

steipete merged commit 70df2b8 into openclaw:main May 13, 2026
22 of 23 checks passed

karmafeast mentioned this pull request May 13, 2026

WO-DRIFT-CURE-20260513: cure upstream drift on PR #79925 head SHA against current upstream/main karmaterminal/openclaw#670

Open

github-actions Bot mentioned this pull request May 13, 2026

📡 Upstream Digest — 2026-05-13 14:10 UTC curtismercier/openclaw-mods#846

Open

This was referenced May 13, 2026

feat: expose rawBody on user messages in plugin hook events #52664

Open

fix: Codex app-server steering path regression #74699

Closed

This was referenced May 14, 2026

Fix external channel follow-up transcript continuity #76941

Open

[Bug]: There is a 3–6-minute delay after sending telegram messages. #78079

Open

brokemac79 mentioned this pull request May 14, 2026

[Bug]: Telegram forum topic messages are serialized across topics before reaching agent pipeline #81953

Closed

clawsweeper Bot mentioned this pull request May 18, 2026

[Bug]: Agent replies to previous message instead of current message (session context confusion) #32296

Open

coygeek mentioned this pull request May 19, 2026

Docs site command queue page conflicts with current active-run steering behavior #83869

Closed

clawsweeper Bot mentioned this pull request May 24, 2026

[Bug]: Feishu plugin: per-chat serial queue prevents messages.queue.mode = "collect" from batching queued messages #54409

Open

This was referenced Jun 3, 2026

fix(tui): respect queue mode in sendMessage busy guard #89078

Open

Feature request: Mid-stream message injection (soft steer) #10960

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: steer mid-turn prompts by default#77023

feat: steer mid-turn prompts by default#77023
steipete merged 17 commits into
openclaw:mainfrom
fuller-stack-dev:codex/default-midturn-steering

fuller-stack-dev commented May 4, 2026 •

edited by steipete

Loading

Uh oh!

clawsweeper Bot commented May 4, 2026 •

edited

Loading

Uh oh!

akrimm702 commented May 13, 2026 •

edited

Loading

Uh oh!

steipete commented May 13, 2026 •

edited

Loading

Uh oh!

steipete commented May 13, 2026 •

edited by clawsweeper Bot

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

fuller-stack-dev commented May 4, 2026 • edited by steipete Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Before

After

Verification

Uh oh!

clawsweeper Bot commented May 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

akrimm702 commented May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

steipete commented May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

steipete commented May 13, 2026 • edited by clawsweeper Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

fuller-stack-dev commented May 4, 2026 •

edited by steipete

Loading

clawsweeper Bot commented May 4, 2026 •

edited

Loading

akrimm702 commented May 13, 2026 •

edited

Loading

steipete commented May 13, 2026 •

edited

Loading

steipete commented May 13, 2026 •

edited by clawsweeper Bot

Loading