Implement executor mesh support with Inbox, OnEvent, and ShouldStop by intel352 · Pull Request #4 · GoCodeAlone/workflow-plugin-agent

intel352 · 2026-04-05T03:25:52Z

Extends executor.Config with Inbox (message injection), OnEvent (real-time streaming), and ShouldStop (custom termination) to support ratchet-cli's agent mesh. 6-step plan, all backward-compatible. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Implement executor mesh support: Inbox, OnEvent, ShouldStop Add three optional fields to executor.Config for mesh support: - Inbox: channel for injecting external messages into the conversation - OnEvent: callback for real-time event streaming - ShouldStop: custom termination hook All features are backward-compatible - nil values preserve existing behavior. Includes comprehensive unit and integration tests. Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/443ef504-27cc-40a0-88f3-b61e67d82a9f Co-authored-by: intel352 <77607+intel352@users.noreply.github.com> * Address code review feedback: fix comment and optimize closed-channel handling Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/443ef504-27cc-40a0-88f3-b61e67d82a9f Co-authored-by: intel352 <77607+intel352@users.noreply.github.com> * Address review: move inbox drain before compaction, copy tool args, fix drainInbox - Move drainInbox call before NeedsCompaction so injected messages are accounted for in compaction decisions - Copy tc.Arguments via copyArgs() before placing on EventToolCallStart to prevent OnEvent handlers from mutating the map used by tool execution - Fix drainInbox: remove dead `closed` variable (default branch now returns false directly), update doc comment to match actual behavior - Add TestExecute_ToolArgsEventIsCopy to verify arg isolation Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/79a383c2-cc77-4f96-8c82-963af9a41e07 Co-authored-by: intel352 <77607+intel352@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: intel352 <77607+intel352@users.noreply.github.com>

Copilot

Pull request overview

This PR adds “executor mesh” features to enable external message injection, real-time event streaming, and custom termination conditions for agents running under a multi-agent mesh.

Changes:

Extended executor.Config with Inbox, OnEvent, and ShouldStop, and integrated these into the main Execute() loop.
Introduced an executor.Event model (executor/events.go) and emitted events for iteration/thinking/text/tool execution/completion/failure.
Added unit + integration-style tests covering inbox draining, event sequencing, termination behavior, and transcript recording.

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
`executor/executor.go`	Implements inbox draining, event emission, and `ShouldStop` termination inside the executor loop.
`executor/events.go`	Defines event types and the `Event` payload used by `OnEvent`.
`executor/executor_test.go`	Adds focused unit tests for inbox/events/ShouldStop/backward compatibility.
`executor/mesh_support_test.go`	Adds end-to-end style mesh scenario tests (inbox + events + termination).
`docs/plans/2026-04-04-executor-mesh-support-design.md`	Documents the design; currently has an ordering mismatch vs implementation for inbox draining vs compaction.
`docs/plans/2026-04-04-executor-mesh-support-plan.md`	Implementation plan; also has the same ordering mismatch.
`go.mod` / `go.sum`	Updates dependencies required by the broader mesh-related changes/tests.

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

Copilot · 2026-04-05T03:29:29Z

+				return &Result{
+					Content:    reason,
+					Status:     "completed",
+					Iterations: iterCount,


The ShouldStop early-return sets Status: "completed" but drops Result.Thinking, unlike the normal completion path (which returns finalThinking). Consider populating Thinking: finalThinking here as well so callers get consistent debugging information for completed runs.

Suggested change

Iterations: iterCount,

Iterations: iterCount,

Thinking: finalThinking,

Copilot · 2026-04-05T03:29:30Z

+	// Emit completed event
+	emit(cfg, Event{
+		Type:      EventCompleted,
+		AgentID:   agentID,
+		Iteration: iterCount,
+		Content:   finalContent,
+	})


Terminal events are only emitted on the normal completion path (EventCompleted at the end) and on Provider.Chat errors (EventFailed). Other early-exit statuses returned from this function (e.g. approval timeout, human request expiry, loop detection) still return without emitting a terminal event, which can leave mesh/TUI consumers waiting for a final signal. Consider emitting a terminal event in every return path (or centralizing returns via a helper that always emits).

intel352 · 2026-04-05T03:30:18Z

@copilot apply changes based on the comments in this thread

…rminal events on all exits, fix docs ordering Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/568b084c-b021-4953-b847-6856a7d13916 Co-authored-by: intel352 <77607+intel352@users.noreply.github.com>

Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/568b084c-b021-4953-b847-6856a7d13916 Co-authored-by: intel352 <77607+intel352@users.noreply.github.com>

Copilot · 2026-04-05T03:38:50Z

@copilot apply changes based on the comments in this thread

All five review comments have been addressed in commits 2db7d2f and e23d510:

emit panic safety — emit now wraps the OnEvent callback in a defer/recover block. Any panic from an external handler is caught and logged via log.Printf so observers can't crash the executor.
ShouldStop missing Thinking — Added Thinking: finalThinking to the early-return Result so callers get consistent debugging info on all completed exits.
Terminal events on all early-exit paths — EventFailed is now emitted before every non-normal return: approval_timeout, request_expired, and loop_detected.
Doc ordering fix (design.md) — Updated the description to say inbox draining happens before the compaction check (matching the implementation).
Plan ordering fix (plan.md) — Aligned step 3 text to match the actual ordering (drain inbox → compaction check → provider.Chat()).

New tests added: TestExecute_OnEventPanicDoesNotCrashExecutor and TestExecute_ShouldStopThinkingPopulated.

Copilot

Pull request overview

Copilot reviewed 7 out of 8 changed files in this pull request and generated no new comments.

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

intel352 · 2026-04-05T03:50:20Z

@copilot Fix the test failures

--- FAIL: TestAssetNameForPlatform_AllPlatforms (0.00s)
    llama_cpp_download_test.go:50: assetNameForPlatform("darwin", "arm64"): want "-bin-macos-arm64.zip", got "-bin-macos-arm64.tar.gz"
    llama_cpp_download_test.go:50: assetNameForPlatform("darwin", "amd64"): want "-bin-macos-x64.zip", got "-bin-macos-x64.tar.gz"
    llama_cpp_download_test.go:50: assetNameForPlatform("linux", "amd64"): want "-bin-linux-amd64.zip", got "-bin-linux-amd64.tar.gz"
    llama_cpp_download_test.go:50: assetNameForPlatform("linux", "arm64"): want "-bin-linux-arm64.zip", got "-bin-linux-arm64.tar.gz"

…tching assetSuffixesForPlatform priority Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/f77f67f5-0445-4eea-a346-636e4fdd9eef Co-authored-by: intel352 <77607+intel352@users.noreply.github.com>

Copilot · 2026-04-05T03:53:10Z

@copilot Fix the test failures

...

Fixed in commit a92a3e7. The test expectations in TestAssetNameForPlatform_AllPlatforms were out of sync with assetSuffixesForPlatform, which prioritizes .tar.gz over .zip for newer llama.cpp releases on non-Windows platforms. Updated the expected values for darwin and linux to .tar.gz.

Copilot

Pull request overview

Copilot reviewed 8 out of 9 changed files in this pull request and generated 3 comments.

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

Copilot · 2026-04-05T03:58:24Z

+			// Emit tool call start event with a copy of the arguments to prevent
+			// any OnEvent handler from mutating the map used by the tool execution.
+			emit(cfg, Event{
+				Type:       EventToolCallStart,
+				AgentID:    agentID,
+				Iteration:  iterCount,
+				ToolName:   tc.Name,
+				ToolCallID: tc.ID,
+				ToolArgs:   copyArgs(tc.Arguments),
+			})


The comment says the ToolArgs copy prevents OnEvent handlers from mutating the args used by tool execution, but copyArgs is only a shallow copy. If tc.Arguments contains nested maps/slices, an OnEvent handler can still mutate nested values and affect tool execution. Consider either updating the comment to reflect the shallow-copy limitation, or performing a deep copy for ToolArgs (with an eye on performance).

Copilot · 2026-04-05T03:58:24Z

+	for {
+		select {
+		case msg, ok := <-cfg.Inbox:
+			if !ok {
+				// Channel closed — stop draining.
+				return messages, true
+			}
+			messages = append(messages, msg)
+			_ = cfg.Transcript.Record(ctx, TranscriptEntry{
+				ID:        uuid.New().String(),
+				AgentID:   agentID,
+				TaskID:    cfg.TaskID,
+				ProjectID: cfg.ProjectID,
+				Iteration: iteration,
+				Role:      msg.Role,
+				Content:   msg.Content,
+			})
+		default:
+			// No more pending messages.
+			return messages, false
+		}


drainInbox can starve the executor loop: if another goroutine is continuously sending to cfg.Inbox, the receive case will keep being ready and this for { select { ... } } loop may never hit default, preventing progress to compaction/LLM calls. Consider bounding the number of messages drained per iteration (e.g., max N) and/or adding a case <-ctx.Done(): to break out early.

Copilot · 2026-04-05T03:58:25Z

+		case msg, ok := <-cfg.Inbox:
+			if !ok {
+				// Channel closed — stop draining.
+				return messages, true
+			}
+			messages = append(messages, msg)
+			_ = cfg.Transcript.Record(ctx, TranscriptEntry{
+				ID:        uuid.New().String(),
+				AgentID:   agentID,
+				TaskID:    cfg.TaskID,
+				ProjectID: cfg.ProjectID,
+				Iteration: iteration,
+				Role:      msg.Role,
+				Content:   msg.Content,
+			})


Inbox messages are appended to the conversation verbatim, including their Role, ToolCallID, and ToolCalls. Since the design/docs describe inbox injection as user-role mesh messages, allowing arbitrary roles here enables accidental or malicious injection of system/assistant/tool messages (prompt injection / transcript corruption). Consider normalizing inbox messages to RoleUser (and clearing tool-related fields) or validating/allowlisting permitted roles before appending/recording.

intel352 · 2026-04-05T04:00:56Z

@copilot apply changes based on the comments in this thread

…fix, inbox role normalization Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/94d3766c-9e42-4886-8dd9-9cc92d2cc3e4 Co-authored-by: intel352 <77607+intel352@users.noreply.github.com>

…t to cover ToolCalls clearing Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/94d3766c-9e42-4886-8dd9-9cc92d2cc3e4 Co-authored-by: intel352 <77607+intel352@users.noreply.github.com>

Copilot · 2026-04-05T04:08:48Z

@copilot apply changes based on the comments in this thread

All three review points addressed in commit 891af7f:

copyArgs comment — Updated to explicitly state this is a shallow copy and that nested values are still shared, with rationale for the trade-off.
drainInbox starvation — Added a drainInboxMaxMessages constant (100) to bound the loop per iteration, added a ctx.Err() pre-check before the loop, and kept the case <-ctx.Done(): inside the select.
Inbox role normalization — Inbox messages are now normalized to RoleUser with ToolCallID and ToolCalls cleared before being appended to the conversation. TestExecute_InboxNormalizesRole covers all four cases (system, assistant, tool, and assistant-with-tool-calls messages).

intel352 and others added 2 commits April 4, 2026 22:43

Copilot AI review requested due to automatic review settings April 5, 2026 03:25

Copilot started reviewing on behalf of intel352 April 5, 2026 03:26 View session

Copilot AI reviewed Apr 5, 2026

View reviewed changes

Copilot started work on behalf of intel352 April 5, 2026 03:30 View session

Copilot AI and others added 2 commits April 5, 2026 03:34

Log recovered panics in emit for better observability

e23d510

Agent-Logs-Url: https://github.com/GoCodeAlone/workflow-plugin-agent/sessions/568b084c-b021-4953-b847-6856a7d13916 Co-authored-by: intel352 <77607+intel352@users.noreply.github.com>

Copilot AI review requested due to automatic review settings April 5, 2026 03:38

intel352 review requested due to automatic review settings April 5, 2026 03:38

Copilot finished work on behalf of intel352 April 5, 2026 03:39

intel352 requested a review from Copilot April 5, 2026 03:41

Copilot started reviewing on behalf of intel352 April 5, 2026 03:42 View session

Copilot AI reviewed Apr 5, 2026

View reviewed changes

Copilot started work on behalf of intel352 April 5, 2026 03:50 View session

Copilot finished work on behalf of intel352 April 5, 2026 03:53

intel352 requested a review from Copilot April 5, 2026 03:54

Copilot started reviewing on behalf of intel352 April 5, 2026 03:55 View session

Copilot AI reviewed Apr 5, 2026

View reviewed changes

Copilot started work on behalf of intel352 April 5, 2026 04:01 View session

Copilot AI and others added 2 commits April 5, 2026 04:04

Copilot AI review requested due to automatic review settings April 5, 2026 04:08

intel352 review requested due to automatic review settings April 5, 2026 04:08

Copilot finished work on behalf of intel352 April 5, 2026 04:09

intel352 merged commit 6b88ebb into master Apr 5, 2026
4 checks passed

intel352 deleted the feat/executor-mesh-support branch April 5, 2026 04:14

	Iterations: iterCount,
	Iterations: iterCount,
	Thinking: finalThinking,

Conversation

intel352 commented Apr 5, 2026 • edited by Copilot AI Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

intel352 commented Apr 5, 2026

Uh oh!

Copilot AI commented Apr 5, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

intel352 commented Apr 5, 2026

Uh oh!

Copilot AI commented Apr 5, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

intel352 commented Apr 5, 2026

Uh oh!

Copilot AI commented Apr 5, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

intel352 commented Apr 5, 2026 •

edited by Copilot AI

Loading