Fix duplicate LLM responses for matched event routines by henrypark133 · Pull Request #1275 · nearai/ironclaw

henrypark133 · 2026-03-17T02:52:22Z

Summary

consume matched event-triggered user messages before they enter the normal chat pipeline
prevent Telegram and gateway-style inbound matches from producing both a routine run and a separate assistant reply
update routine event trace coverage to assert the matched message only produces the routine-driven LLM turn

Testing

cargo test --features libsql routine_event_trigger_

gemini-code-assist · 2026-03-17T02:52:42Z

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request resolves an issue causing duplicate LLM responses when event-triggered routines are activated by inbound user messages. By introducing an early check for routine triggers and consuming the message if a routine fires, the system now ensures that only the routine-driven response is generated. This change streamlines the interaction flow, prevents redundant replies from the main agent, and improves the overall consistency of the agent's behavior.

Highlights

Early Message Consumption: Implemented logic to consume user input messages early in the agent loop if they trigger an event-driven routine, preventing further processing by the main agent.
Duplicate Response Prevention: Eliminated scenarios where both the main agent and an event-triggered routine would respond to the same inbound message, ensuring a single, consistent LLM turn.
Test Coverage Enhancement: Updated end-to-end tests to specifically assert that only a single LLM call and response occur when an event-triggered routine is matched, validating the fix.

Changelog

src/agent/agent_loop.rs
- Added a new conditional block to Agent::run_loop that checks for event-triggered routines on non-internal user input messages. If a routine fires, the message is consumed, and the loop continues to the next message.
- Removed the previous, general check_event_triggers call that occurred later in the Agent::run_loop method.
tests/e2e_advanced_traces.rs
- Introduced rig.clear().await and llm_calls_before tracking before sending incoming messages to ensure accurate LLM call counts.
- Added assertions to verify that llm_call_count() increases by exactly one after an event-triggered routine fires, confirming only the routine's LLM call.
- Modified rig.wait_for_responses to expect only one response and added an assertion to confirm responses.len() is 1, preventing multiple responses.
- Updated an assertion to check for specific content in the single expected routine notification.
tests/fixtures/llm_traces/advanced/routine_event_any_channel.json
- Removed a redundant LLM response entry, reflecting the fix for duplicate responses.
tests/fixtures/llm_traces/advanced/routine_event_telegram.json
- Removed a redundant LLM response entry, reflecting the fix for duplicate responses.

Activity

The author, henrypark133, has provided detailed testing instructions, including a specific cargo test command to verify the fix: cargo test --features libsql routine_event_trigger_.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request addresses an issue where messages triggering event-based routines would also be processed by the main chat pipeline, resulting in duplicate responses. The fix is implemented by checking for and consuming event-triggered user messages before they enter the standard message handling logic. If a routine is fired, the agent loop continues to the next message, preventing a duplicate agent reply. The end-to-end tests and corresponding LLM trace fixtures have been updated to assert that only the routine-driven LLM turn occurs for a matched message, confirming the fix.

Copilot

Pull request overview

This PR fixes a duplicate-response behavior where a single inbound message that matches an event-triggered routine could both (1) fire the routine and (2) still flow through the normal agent chat pipeline, producing an extra assistant reply.

Changes:

Consume inbound event-matching user messages earlier to prevent the main chat/tool pipeline from generating a second assistant response.
Update E2E trace fixtures to remove the extra (duplicate) assistant response step.
Strengthen E2E assertions to verify only the routine-driven LLM call occurs and only the routine notification is emitted.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
`src/agent/agent_loop.rs`	Adds early event-trigger checking and skips the normal message handling when an event routine fires, aiming to prevent duplicate responses.
`tests/e2e_advanced_traces.rs`	Updates event-trigger tests to assert exactly one additional LLM call and a single routine notification response after a matching inbound message.
`tests/fixtures/llm_traces/advanced/routine_event_telegram.json`	Removes the previously-recorded extra assistant response from the trace fixture.
`tests/fixtures/llm_traces/advanced/routine_event_any_channel.json`	Removes the previously-recorded extra assistant response from the trace fixture.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot

Pull request overview

This PR fixes a behavior where an inbound message that matches an event-triggered routine could produce both (1) a routine-driven LLM turn/notification and (2) a separate assistant reply from the normal chat pipeline. It does this by consuming matched event-triggered user messages during handle_message() so they don’t proceed through the standard response path, and updates trace fixtures/tests accordingly.

Changes:

Move event-trigger checking into handle_message() and suppress the normal assistant reply by returning an empty response when event routines fire.
Add LLM-call-count and response-count assertions to ensure a matched event message produces only the routine-driven LLM call/notification.
Update LLM trace fixtures to remove the previously duplicated assistant reply.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
`src/agent/agent_loop.rs`	Consumes matched event-trigger messages inside `handle_message()` by firing routines and returning an empty response to prevent duplicate assistant replies.
`tests/e2e_advanced_traces.rs`	Strengthens E2E assertions to ensure only one LLM call and one routine notification occur for matched event messages.
`tests/fixtures/llm_traces/advanced/routine_event_telegram.json`	Removes the extra assistant response from the expected trace for Telegram-scoped event routines.
`tests/fixtures/llm_traces/advanced/routine_event_any_channel.json`	Removes the extra assistant response from the expected trace for any-channel event routines.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

henrypark133 · 2026-03-18T16:58:48Z

+        if matches!(submission, Submission::UserInput { .. })
+            && let Some(engine) = self.routine_engine().await
+        {
+            let fired = engine.check_event_triggers(message).await;
+            if fired > 0 {
+                tracing::debug!(
+                    channel = %message.channel,
+                    user = %message.user_id,
+                    fired,
+                    "Consumed inbound user message with matching event-triggered routine(s)"
+                );
+                return Ok(Some(String::new()));
+            }


Fixed in 89320c0. routine_trigger_message() now extracts the post-hook content from the parsed Submission::UserInput { content } and, when it differs from the original message.content, clones the message with the rewritten content (via Cow::Owned). When content is unchanged, Cow::Borrowed avoids any allocation. The trigger check now consistently matches against what the agent will actually process.

@@ -191,6 +191,11 @@ impl Agent {
        self.routine_engine_slot = Some(slot);
    }

+    async fn routine_engine(&self) -> Option<Arc<crate::agent::routine_engine::RoutineEngine>> {
+        let slot = self.routine_engine_slot.as_ref()?;
+        slot.read().await.clone()
+    }


zmanian

Review: fix duplicate LLM responses for event-triggered routines

Clean, well-structured fix. The progression through 4 commits shows good iteration:

Initial fix (consume before handle_message)
Rustfmt
Move check inside handle_message to preserve preprocessing
Match against rewritten input using Cow for efficiency

The final approach is correct -- checking event triggers inside handle_message (after submission parsing) ensures routines fire against preprocessed content and that the message is consumed before the main chat pipeline processes it.

Positives:

routine_trigger_message() helper uses Cow nicely to avoid cloning when content is unchanged
Test coverage is solid: unit tests for the helper (borrowed, owned, internal skip), updated integration tests asserting only 1 response and 1 LLM call, fixture traces updated
The routine_engine_slot initialization simplification is clean
CI fully green

Minor notes:

Returning Ok(Some(String::new())) relies on the caller's !response.is_empty() guard to skip empty responses -- this coupling is correct but implicit. A future refactor might consider Ok(None) for "consumed, no response needed" but that's outside scope.

LGTM.

Copilot

Pull request overview

This PR prevents duplicate assistant turns when an inbound user message matches an event-triggered routine by consuming the message inside handle_message() (after standard preprocessing) and returning an empty response.

Changes:

Adds a helper to derive the routine-trigger “effective message” from the parsed Submission (including hook-modified user input).
Moves event-trigger consumption into handle_message() and suppresses the normal assistant reply when routines fire.
Adds unit tests validating the trigger-message borrowing/rewriting behavior and internal-message exclusion.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

…_slot Address Copilot review feedback: - Change check_event_triggers to accept (user_id, channel, content) instead of &IncomingMessage, eliminating the need to clone the full message (including attachments) when hooks rewrite content. - Remove routine_trigger_message and the Cow<IncomingMessage> indirection; the event-trigger check now inlines the is_internal + UserInput guard and passes the post-hook content string directly. - Make routine_engine_slot non-optional since Agent::new() always initializes it. Removes the redundant Option wrapper and simplifies accessor/setter methods. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: consume matched event routine messages * style: run rustfmt for event routine fix * fix: preserve preprocessing for routine-triggered messages * fix: match routines against rewritten input * refactor: narrow check_event_triggers API and simplify routine_engine_slot Address Copilot review feedback: - Change check_event_triggers to accept (user_id, channel, content) instead of &IncomingMessage, eliminating the need to clone the full message (including attachments) when hooks rewrite content. - Remove routine_trigger_message and the Cow<IncomingMessage> indirection; the event-trigger check now inlines the is_internal + UserInput guard and passes the post-hook content string directly. - Make routine_engine_slot non-optional since Agent::new() always initializes it. Removes the redundant Option wrapper and simplifies accessor/setter methods. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Copilot AI review requested due to automatic review settings March 17, 2026 02:52

github-actions Bot added scope: agent Agent core (agent loop, router, scheduler) size: M 50-199 changed lines risk: medium Business logic, config, or moderate-risk modules contributor: core 20+ merged PRs labels Mar 17, 2026

Copilot started reviewing on behalf of henrypark133 March 17, 2026 02:52 View session

gemini-code-assist Bot reviewed Mar 17, 2026

View reviewed changes

Copilot AI reviewed Mar 17, 2026

View reviewed changes

Comment thread src/agent/agent_loop.rs Outdated

Comment thread src/agent/agent_loop.rs Outdated

henrypark133 requested a review from Copilot March 17, 2026 03:07

Copilot started reviewing on behalf of henrypark133 March 17, 2026 03:08 View session

Copilot AI reviewed Mar 17, 2026

View reviewed changes

zmanian previously approved these changes Mar 17, 2026

View reviewed changes

henrypark133 mentioned this pull request Mar 17, 2026

Routines: preserve no-duplicate-response behavior for matched event triggers #1323

Closed

henrypark133 added 4 commits March 18, 2026 09:13

fix: consume matched event routine messages

e203b1a

style: run rustfmt for event routine fix

5452126

fix: preserve preprocessing for routine-triggered messages

0d0e0f1

fix: match routines against rewritten input

89320c0

Copilot AI review requested due to automatic review settings March 18, 2026 16:16

henrypark133 dismissed zmanian’s stale review via 89320c0 March 18, 2026 16:16

henrypark133 force-pushed the codex/routine-trigger-fix branch from 18a6fc2 to 89320c0 Compare March 18, 2026 16:16

Copilot started reviewing on behalf of henrypark133 March 18, 2026 16:16 View session

Copilot AI reviewed Mar 18, 2026

View reviewed changes

Comment thread src/agent/agent_loop.rs Outdated

Comment thread src/agent/agent_loop.rs Outdated

nickpismenkov approved these changes Mar 18, 2026

View reviewed changes

henrypark133 merged commit 2020270 into staging Mar 18, 2026
14 checks passed

henrypark133 deleted the codex/routine-trigger-fix branch March 18, 2026 19:29

BrewTestBot mentioned this pull request Mar 19, 2026

ironclaw 0.20.0 Homebrew/homebrew-core#273275

Merged

Navigation Menu

Fix duplicate LLM responses for matched event routines - #1275

Fix duplicate LLM responses for matched event routines#1275
henrypark133 merged 5 commits into
stagingfrom
codex/routine-trigger-fix

henrypark133 commented Mar 17, 2026

Uh oh!

gemini-code-assist Bot commented Mar 17, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

henrypark133 Mar 18, 2026

Uh oh!

zmanian left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

henrypark133 commented Mar 17, 2026

Summary

Testing

Uh oh!

gemini-code-assist Bot commented Mar 17, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

henrypark133 Mar 18, 2026

Choose a reason for hiding this comment

Uh oh!

zmanian left a comment

Choose a reason for hiding this comment

Review: fix duplicate LLM responses for event-triggered routines

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants