fix(channels): strip silence tokens mixed with reply text#164
Merged
Conversation
Every channel bridge gated the "agent stayed silent" path on `responseText.trim() === NO_REPLY_TAG`. That only matched when the model emitted the token alone — but models occasionally emit a real reply AND the token in the same final string (e.g. "ok.\n<NO_REPLY>"), in which case the literal "<NO_REPLY>" leaked through to Telegram / Discord / Slack / WeChat / Signal. Promote the silence-token logic to @openhermit/shared as stripSilenceTokens: removes every occurrence of <NO_REPLY> / <EMPTY_RESPONSE> from anywhere in the text and reports both the remainder and whether the result is effectively silent. Each bridge now drops the response only when the *stripped* remainder is empty, and sends/edits with the cleaned text when the model mixed both forms. Includes a node:test suite for the helper covering the regression case.
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (3)
🚧 Files skipped from review as they are similar to previous changes (3)
📝 WalkthroughWalkthroughAdds a shared stripSilenceTokens utility and updates all channel bridges to use it for streaming edits and final-response suppression, replacing per-bridge NO_REPLY sentinel logic. ChangesSilence token implementation and bridge migration
Sequence DiagramsequenceDiagram
participant Agent
participant Bridge
participant StripFn as stripSilenceTokens
participant ChannelAPI as Channel API
Agent->>Bridge: final accumulated response
Bridge->>StripFn: rawResponseText
StripFn-->>Bridge: {text, hadToken, isSilent}
alt isSilent == true
Bridge->>Channel API: delete provisional message
Bridge-->>Agent: {text: undefined, error: undefined}
else isSilent == false
alt hadToken == true
Bridge->>Channel API: send/edit with stripped text
else hadToken == false
Bridge->>Channel API: send/edit with original text
end
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Warning There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure. 🔧 ESLint
ESLint skipped: no ESLint configuration detected in root package.json. To enable, add Comment |
There was a problem hiding this comment.
Pull request overview
This PR fixes a production regression where the agent’s “silence” sentinel tokens (e.g. <NO_REPLY>) could be emitted alongside real reply text and end up displayed in downstream channel messages. It centralizes token-stripping logic in @openhermit/shared and updates channel bridges to suppress replies only when the post-stripped remainder is empty, otherwise sending the cleaned text.
Changes:
- Add
stripSilenceTokens(andSILENCE_TOKENS) to@openhermit/sharedand export it from the package entrypoint. - Update Slack/Discord/Telegram/Signal/WeChat bridges to use the shared stripping logic instead of strict equality checks.
- Add a new
node:testsuite covering bare-token silence, mixed-content stripping, multiple occurrences, and legacy token handling.
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 5 comments.
Show a summary per file
| File | Description |
|---|---|
| packages/shared/src/silence-tokens.ts | Introduces the shared token-stripping helper and result contract. |
| packages/shared/src/index.ts | Re-exports the new helper/types from @openhermit/shared. |
| packages/shared/test/silence-tokens.test.ts | Adds regression and behavior tests for stripping/silence detection. |
| apps/channels/slack/src/bridge.ts | Uses stripSilenceTokens to prevent silence tokens leaking into Slack messages. |
| apps/channels/discord/src/bridge.ts | Uses stripSilenceTokens to prevent silence tokens leaking into Discord messages. |
| apps/channels/telegram/src/bridge.ts | Uses stripSilenceTokens to prevent silence tokens leaking into Telegram messages (incl. delete-on-silence). |
| apps/channels/signal/src/bridge.ts | Uses stripSilenceTokens so Signal suppresses token-only responses and strips mixed tokens. |
| apps/channels/wechat/src/bridge.ts | Uses stripSilenceTokens before sending outbound WeChat messages. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
- Strip silence tokens from streaming text too: slack/discord/telegram all edit partial messages with accumulatedText during text_delta. If <NO_REPLY> arrived mid-stream the token would flash visibly until the final edit. Compute displayText once per delta and route it through every streaming send/edit call. - Wire packages/shared into the root test glob and add a per-package test script so the new node:test suite runs in CI. - Clarify isSilent docstring to mirror the consumer-side definition. - Update docs/channel-adapter.md and docs/session-model.md so the documented NO_REPLY contract reflects strip-anywhere semantics.
Conflicts: apps/channels/telegram/src/bridge.ts — kept local removal of NO_REPLY_TAG, kept main's new shouldSpeak/VOICE_MAX_TEXT_LENGTH helpers. Also applies the same silence-token-strip fix to apps/channels/whatsapp (the new Baileys adapter from #155), so the WhatsApp bridge doesn't ship with the brittle equality check the rest of this PR removes.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/session-model.md`:
- Line 86: The docs sentence that currently references only `<NO_REPLY>` should
be updated to reflect runtime behavior by mentioning both `<NO_REPLY>` and the
legacy `<EMPTY_RESPONSE>` markers; change the text that says "`<NO_REPLY>`
markers are stripped..." to something like "Both `<NO_REPLY>` and legacy
`<EMPTY_RESPONSE>` markers are stripped by channel adapters; if the stripped
remainder is empty the reply is suppressed entirely (whether the model emitted
the token alone or alongside real text)" so the documentation matches the
implementation.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 6e1732a7-8c9b-4d83-945b-799254860f56
⛔ Files ignored due to path filters (1)
package-lock.jsonis excluded by!**/package-lock.json
📒 Files selected for processing (9)
apps/channels/discord/src/bridge.tsapps/channels/slack/src/bridge.tsapps/channels/telegram/src/bridge.tsapps/channels/whatsapp/src/bridge.tsdocs/channel-adapter.mddocs/session-model.mdpackage.jsonpackages/shared/package.jsonpackages/shared/src/silence-tokens.ts
✅ Files skipped from review due to trivial changes (2)
- docs/channel-adapter.md
- package.json
🚧 Files skipped from review as they are similar to previous changes (3)
- apps/channels/slack/src/bridge.ts
- packages/shared/src/silence-tokens.ts
- apps/channels/telegram/src/bridge.ts
session-model.md mentioned only <NO_REPLY> but the runtime strips both <NO_REPLY> and the legacy <EMPTY_RESPONSE>. channel-adapter.md already called both out — bring session-model.md in line.
|
Actionable comments posted: 0 |
This was referenced May 29, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary by CodeRabbit