Skip to content

fix(auto-reply): restrict NO_REPLY to automatic group replies#82270

Merged
steipete merged 3 commits into
mainfrom
fix/webchat-direct-no-reply-prompt-82254
May 15, 2026
Merged

fix(auto-reply): restrict NO_REPLY to automatic group replies#82270
steipete merged 3 commits into
mainfrom
fix/webchat-direct-no-reply-prompt-82254

Conversation

@steipete

@steipete steipete commented May 15, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Summary:

  • Restricts NO_REPLY prompt guidance to legacy automatic group/channel replies only; direct chats and message_tool_only source turns stay free of silent-token instructions.
  • Removes legacy silentReplyRewrite config/runtime fallback behavior, including dead pending-spawn rewrite plumbing, and adds doctor cleanup for old rewrite/direct silent-reply keys.
  • Suppresses accidental exact silent tokens during direct/routed delivery instead of rewriting them into visible fallback text; docs, changelog, generated config baseline, and delivery tests updated.

Verification:

  • pnpm docs:list
  • pnpm config:schema:gen && pnpm config:docs:gen
  • node --import tsx scripts/generate-config-doc-baseline.ts --check
  • node scripts/run-vitest.mjs src/infra/outbound/payloads.test.ts src/commands/doctor/shared/legacy-config-migrate.test.ts src/auto-reply/reply/groups.test.ts src/agents/prompt-composition.test.ts src/auto-reply/reply/route-reply.test.ts - 5 files, 109 tests passed
  • node scripts/run-vitest.mjs src/shared/silent-reply-policy.test.ts src/config/silent-reply.test.ts src/auto-reply/reply/groups.test.ts src/agents/prompt-composition.test.ts src/infra/outbound/payloads.test.ts src/infra/outbound/deliver.test.ts src/auto-reply/reply/reply-flow.test.ts src/auto-reply/reply/route-reply.test.ts src/auto-reply/reply/get-reply-run.media-only.test.ts src/commands/doctor/shared/legacy-config-migrate.test.ts extensions/telegram/src/bot/delivery.test.ts extensions/telegram/src/bot-message-dispatch.test.ts - 12 files, 366 tests passed
  • node scripts/run-vitest.mjs src/agents/prompt-composition.test.ts src/auto-reply/reply/route-reply.test.ts - 2 files, 32 tests passed
  • node scripts/run-vitest.mjs src/config/zod-schema.test.ts src/config/zod-schema.agent-defaults.test.ts src/config/schema.help.test.ts - 1 file found, 25 tests passed
  • pnpm tsgo:core --noEmit --pretty false
  • git diff --check
  • Prompt snapshot grep: no NO_REPLY, Silent Replies, respond with ONLY, or reply with exactly in direct/message-tool-only snapshots
  • codex-review --mode auto --full-access --output /tmp/codex-review-82254-cleanup-final.txt - clean, no accepted/actionable findings; review also ran 3 focused files, 77 tests, plus config docs baseline check
  • pnpm check:changed delegated to Testbox tbx_01krphwn6jkcdvp1wmqbnsj2x1; Blacksmith stayed queued for ~9m with no worker/IP/run URL, then the local waiter was stopped (exit -1), so no remote command executed

Behavior addressed: NO_REPLY is now only prompted for automatic group/channel replies where silence is allowed. Direct chats and message-tool-only source turns are quiet by not sending, not by asking the model to emit NO_REPLY.
Real environment tested: local repo unit/config/docs/type checks plus Codex review; Testbox changed gate was attempted as tbx_01krphwn6jkcdvp1wmqbnsj2x1 but stayed queued with no worker/IP/run URL and was stopped.
Exact steps or command run after this patch: focused Vitest files above, pnpm tsgo:core --noEmit --pretty false, config doc generation/check, prompt snapshot grep, git diff --check, pnpm docs:list, Codex review.
Evidence after fix: direct prompt invariant asserts no NO_REPLY; group automatic prompt still asserts exact NO_REPLY; message_tool_only prompt snapshots contain no NO_REPLY; dispatcher/outbound/Telegram route exact silent tokens to no delivery instead of rewrite text; generated config baseline no longer lists removed rewrite/direct keys.
Observed result after fix: direct and message-tool-only surfaces receive no silent-token system prompt; legacy group auto mode keeps the exact silent-token instruction; old silentReplyRewrite config is removed by doctor; dead rewrite-only pending-spawn hook is gone.
What was not tested: prompt snapshot generator/check hung before producing output and was killed; Testbox pnpm check:changed never got a worker and did not execute remotely; no live Telegram send was run for this refactor pass.

Fixes #82254

@openclaw-barnacle openclaw-barnacle Bot added agents Agent runtime and tooling size: XS maintainer Maintainer-authored PR labels May 15, 2026
@clawsweeper

clawsweeper Bot commented May 15, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Codex review: needs maintainer review before merge.

Summary
The PR removes direct/message-tool-only NO_REPLY prompt guidance, drops legacy silent-reply rewrite config/runtime paths, suppresses exact silent-token deliveries, and updates doctor migration, docs, generated config baselines, changelog, and tests.

Reproducibility: no. not as a live end-to-end reproduction in this read-only review. Source and v2026.5.12 inspection show the direct-chat path can receive exact NO_REPLY guidance while direct rewrite is enabled by default.

Real behavior proof
Override: A maintainer applied proof: override for this PR.

Next step before merge
Protected maintainer-labeled PR with a config-contract removal and merge-validation gaps; there is no narrow automated repair to queue from this review.

Security
Cleared: No concrete security or supply-chain concern was found in the diff.

Review details

Best possible solution:

Land this only after maintainers accept the config-contract removal and the PR head has appropriate changed-gate coverage, with the prompt/delivery fix kept scoped to preventing direct and message-tool-only silent-token leakage.

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

No, not as a live end-to-end reproduction in this read-only review. Source and v2026.5.12 inspection show the direct-chat path can receive exact NO_REPLY guidance while direct rewrite is enabled by default.

Is this the best way to solve the issue?

Partly yes: removing direct/message-tool-only silent-token guidance and suppressing exact accidental tokens is the narrow code-path fix. The legacy config removal is broader and needs maintainer product/API acceptance before merge.

What I checked:

  • Protected PR labels: The provided live PR context lists maintainer, proof: override, channel: telegram, and mantis: telegram-visible-proof, so cleanup policy keeps the PR open for explicit maintainer handling while treating proof as overridden.
  • Current shipped behavior can inject direct-chat NO_REPLY guidance: v2026.5.12 has default direct silent-reply rewrite enabled and buildDirectChatContext can tell direct chats to reply exactly with NO_REPLY for fallback delivery. (src/auto-reply/reply/groups.ts:309, f066dd2f31c2)
  • Current main still has the source-level prompt path: Current main passes resolved silent policy and rewrite settings into buildDirectChatContext, whose direct-chat branch can emit exact silent-token instructions when rewrite is true. (src/auto-reply/reply/get-reply-run.ts:470, 0b7ff665f664)
  • PR head removes direct prompt and generic messaging token guidance: The PR diff removes silent-policy/rewrite inputs from direct prompt construction and passes silentReplyPromptMode through the Messaging section so direct/message-tool-only turns avoid NO_REPLY guidance. (src/agents/system-prompt.ts:1251, d430ebb23be5)
  • PR head suppresses exact silent-token delivery instead of rewriting: The PR diff removes the dispatcher rewrite/preserve branch and routes exact silent final payloads through normal normalization, which skips bare silent payloads before delivery. (src/auto-reply/reply/reply-dispatcher.ts:154, d430ebb23be5)
  • Config migration coverage is included but needs maintainer acceptance: The PR adds doctor migration rules for removed silentReplyRewrite and direct silent-reply keys, which addresses old configs but also changes the public config contract. (src/commands/doctor/shared/legacy-config-migrations.runtime.agents.ts:95, d430ebb23be5)

Likely related people:

  • steipete: Current-main blame for the direct-chat context, dispatcher rewrite, and outbound silent-reply planning points to Peter Steinberger, and nearby recent commits touched message delivery and channel history paths used by this PR. (role: recent area contributor; confidence: high; commits: 1ba75463a520, 4e6c85d930bd, a4b17d65a8ff; files: src/auto-reply/reply/groups.ts, src/auto-reply/reply/reply-dispatcher.ts, src/infra/outbound/payloads.ts)
  • Takhoffman: Earlier merged commits introduced and refined the silent-reply policy and prompt-alignment behavior that this PR changes. (role: feature-history owner; confidence: medium; commits: 1303b032412b, cc57d56b92f5; files: src/shared/silent-reply-policy.ts, src/config/silent-reply.ts, src/auto-reply/reply/groups.ts)

Remaining risk / open question:

  • The PR body says the Testbox changed gate never executed and no live Telegram send was run; proof: override bypasses the contributor proof gate but does not replace merge validation.
  • Removing silentReplyRewrite and direct silentReply config is a public config-contract change that should be explicitly accepted by maintainers before landing.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 0b7ff665f664.

Re-review progress:

@clawsweeper clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 15, 2026
@openclaw-barnacle openclaw-barnacle Bot added docs Improvements or additions to documentation channel: telegram Channel integration: telegram commands Command implementations size: L and removed size: XS labels May 15, 2026
@steipete steipete changed the title fix(auto-reply): avoid NO_REPLY prompt in direct chats fix(auto-reply): restrict NO_REPLY to automatic group replies May 15, 2026
@clawsweeper clawsweeper Bot added mantis: telegram-visible-proof Mantis should capture Telegram visible proof. and removed proof: sufficient ClawSweeper judged the real behavior proof convincing. labels May 15, 2026
@steipete
steipete force-pushed the fix/webchat-direct-no-reply-prompt-82254 branch from d5743af to d430ebb Compare May 15, 2026 20:19
@steipete steipete added the proof: override Maintainer override for the external PR real behavior proof gate. label May 15, 2026
@steipete
steipete merged commit 445ed9b into main May 15, 2026
88 of 94 checks passed
@steipete
steipete deleted the fix/webchat-direct-no-reply-prompt-82254 branch May 15, 2026 20:29
@steipete

steipete commented May 15, 2026

Copy link
Copy Markdown
Contributor Author

Landed via rebase merge onto main.

  • Gate: rebased onto current origin/main; resolved CHANGELOG.md and generated config baseline conflicts; ran focused Vitest (5 files, 109 tests), node --import tsx scripts/generate-config-doc-baseline.ts --check, pnpm tsgo:core --noEmit --pretty false, and git diff --check on the rebased branch.
  • GitHub: Real behavior proof passed with maintainer proof: override; preflight/security/dependency/doc quick gates passed, broader matrix remained queued and non-blocking when GitHub accepted the merge.
  • Source head: d430ebb
  • Land commit: 445ed9b

Thanks @absol89 for the report.

@steipete steipete mentioned this pull request May 15, 2026
Closed
This was referenced May 15, 2026
Closed
Closed
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

agents Agent runtime and tooling channel: telegram Channel integration: telegram commands Command implementations docs Improvements or additions to documentation maintainer Maintainer-authored PR mantis: telegram-visible-proof Mantis should capture Telegram visible proof. proof: override Maintainer override for the external PR real behavior proof gate. size: XL

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

[Bug]: Agent replies NO! or NO in the openclaw web UI after upgrading from 5.7 to 5.12 and using openclaw doctor

1 participant

@steipete