fix: wire deferred review items end-to-end by Telli · Pull Request #39 · clawdotnet/openclaw.net

Telli · 2026-04-01T05:27:23Z

Summary

Follow-up to #38 addressing review comments that required deeper integration across the runtime.

Multi-agent routing wired: AgentRouteResolver now runs in GatewayWorkers inbound processing, applying model overrides from matched routes to sessions
/think now functional: session.ReasoningEffort plumbed into ChatOptions.AdditionalProperties["reasoning_effort"] for both RunAsync and RunStreamingAsync paths
/compact now functional: CompactHistoryAsync made public, callback wired from gateway initialization — LLM-powered summarization instead of simple trim
/verbose now functional: Appends tool call count and token delta footer to responses when enabled
DmPolicy enforced for Slack/Discord/Signal: Extended policy resolution in GatewayWorkers and validation in ConfigValidator
Roadmap updated: Moved completed channel/tool/preset/command work to "Recently Completed"

Test plan

Full solution builds with zero errors/warnings
All 624 tests pass
Test /think high sets reasoning_effort in LLM call
Test /compact triggers LLM summarization (vs simple trim fallback)
Test /verbose on shows tool call footer in responses
Test DmPolicy=closed drops messages for slack/discord/signal channels
Test routing config applies model override to matched sessions

🤖 Generated with Claude Code

Addresses remaining PR review comments that required deeper integration: Multi-agent routing: - Wire AgentRouteResolver into GatewayWorkers inbound processing - Apply route model overrides to sessions after creation Reasoning effort (/think): - Plumb session.ReasoningEffort into ChatOptions.AdditionalProperties for both RunAsync and RunStreamingAsync LLM call paths History compaction (/compact): - Make CompactHistoryAsync public on AgentRuntime - Wire SetCompactCallback from gateway runtime initialization so /compact triggers LLM-powered summarization instead of simple trim Verbose mode (/verbose): - Append tool call count and token delta footer to responses when session.VerboseMode is enabled DM policy enforcement: - Extend DmPolicy resolution in GatewayWorkers to cover Slack, Discord, and Signal channels (was only SMS/Telegram/WhatsApp/Teams) - Add DmPolicy validation for new channels in ConfigValidator Documentation: - Update ROADMAP.md to move completed channel/tool/preset/command work from future plans to "Recently Completed" Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

augmentcode · 2026-04-01T05:32:22Z

🤖 Augment PR Summary

Summary: Follow-up PR to wire previously deferred review items end-to-end across the gateway runtime.

Changes:

Plumbs /think into LLM calls via ChatOptions.AdditionalProperties["reasoning_effort"] (sync + streaming).
Wires /compact to trigger LLM-powered history compaction from gateway initialization.
Enforces and validates DmPolicy for Slack/Discord/Signal in inbound processing and config validation.
Adds inbound multi-agent route resolution and applies per-route model overrides to sessions.
Adds /verbose response footer with tool-call count and per-turn token deltas.

_{🤖 Was this summary useful? React with 👍 or 👎}

augmentcode

Review completed. 5 suggestions posted.

Comment augment review to trigger a new review at any time.

augmentcode · 2026-04-01T05:32:23Z

src/OpenClaw.Gateway/Composition/RuntimeInitializationExtensions.cs

+        {
+            services.CommandProcessor.SetCompactCallback(async (session, ct) =>
+            {
+                var before = session.History.Count;


before is computed but never used, and the callback returns a derived kept value that may not reflect what was actually preserved when CompactHistoryAsync falls back to TrimHistory. This can make the /compact response message report misleading “recent turns preserved” counts.

Severity: low

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

augmentcode · 2026-04-01T05:32:23Z

src/OpenClaw.Gateway/Extensions/GatewayWorkers.cs

+                            if (resolvedRoute is not null)
+                            {
+                                if (resolvedRoute.ModelOverride is not null)
+                                    session.ModelOverride ??= resolvedRoute.ModelOverride;


Using session.ModelOverride ??= resolvedRoute.ModelOverride means routing won’t update the model for sessions that already have ModelOverride set (including persisted/stale overrides), so routing config changes may not take effect for existing sessions.

Severity: medium

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

augmentcode · 2026-04-01T05:32:23Z

src/OpenClaw.Gateway/Extensions/GatewayWorkers.cs

+                                if (session.VerboseMode)
+                                {
+                                    var lastTurn = session.History.Count > 0 ? session.History[^1] : null;
+                                    var toolCallCount = lastTurn?.ToolCalls?.Count ?? 0;


toolCallCount is derived from session.History[^1].ToolCalls, but tool calls are stored on the separate [tool_use] history turn (the final assistant text turn usually has ToolCalls=null). This will often show 0 tool calls even when tools were invoked.

Severity: medium

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

augmentcode · 2026-04-01T05:32:23Z

src/OpenClaw.Gateway/Extensions/GatewayWorkers.cs

                                    heartbeatService.RecordResult(session, responseText, suppressHeartbeatDelivery, inputTokenDelta, outputTokenDelta);

+                                // Append verbose mode footer (tool calls and token delta)
+                                if (session.VerboseMode)


Verbose mode footer is only appended in the non-streaming branch; websocket envelope streaming responses won’t include the tool/token footer even when /verbose on is set.

Severity: low

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

augmentcode · 2026-04-01T05:32:23Z

src/OpenClaw.Agent/AgentRuntime.cs

                : null
        };

+        if (!string.IsNullOrWhiteSpace(session.ReasoningEffort))


This wires reasoning_effort for the native AgentRuntime, but MafAgentRuntime constructs its own ChatOptions and doesn’t apply session.ReasoningEffort, so /think may be a no-op when Runtime.Orchestrator is MAF.

Severity: medium

Other Locations

src/OpenClaw.Agent/AgentRuntime.cs:401

src/OpenClaw.MicrosoftAgentFrameworkAdapter/MafAgentRuntime.cs:402

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

Copilot

Pull request overview

This PR completes end-to-end wiring for previously added gateway commands and routing/policy behavior, integrating them into the inbound processing pipeline and agent runtime execution paths.

Changes:

Enforces DmPolicy for Slack/Discord/Signal in GatewayWorkers and validates these config fields at startup.
Wires multi-agent route resolution into inbound processing and applies route model overrides onto sessions.
Implements /think, /compact, and /verbose functionality by plumbing ReasoningEffort, wiring an LLM-backed compaction callback, and appending a verbose footer.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/OpenClaw.Gateway/Extensions/GatewayWorkers.cs	Enforces DM policy for additional channels, resolves routes on inbound, applies model override, and appends verbose footer.
src/OpenClaw.Gateway/Composition/RuntimeInitializationExtensions.cs	Wires `/compact` callback to invoke LLM-backed history compaction.
src/OpenClaw.Core/Validation/ConfigValidator.cs	Adds validation for Slack/Discord/Signal `DmPolicy` fields.
src/OpenClaw.Agent/AgentRuntime.cs	Plumbs `reasoning_effort` into LLM options and exposes `CompactHistoryAsync` publicly.
docs/ROADMAP.md	Updates roadmap to move completed work into “Recently Completed”.

Comments suppressed due to low confidence (1)

src/OpenClaw.Gateway/Extensions/GatewayWorkers.cs:381

DmPolicy is validated case-insensitively (see ValidateDmPolicy), but the enforcement here uses case-sensitive pattern matching (policy is "closed" / policy is "pairing"). This can cause "Closed"/"PAIRING" configs to bypass the policy. Normalize policy (e.g., ToLowerInvariant()) or compare with StringComparison.OrdinalIgnoreCase before enforcing.

                            // ── DM Pairing Check ─────────────────────────────────
                            var policy = "open";
                            if (msg.ChannelId == "sms") policy = config.Channels.Sms.DmPolicy;
                            if (msg.ChannelId == "telegram") policy = config.Channels.Telegram.DmPolicy;
                            if (msg.ChannelId == "whatsapp") policy = config.Channels.WhatsApp.DmPolicy;
                            if (msg.ChannelId == "teams") policy = config.Channels.Teams.DmPolicy;
                            if (msg.ChannelId == "slack") policy = config.Channels.Slack.DmPolicy;
                            if (msg.ChannelId == "discord") policy = config.Channels.Discord.DmPolicy;
                            if (msg.ChannelId == "signal") policy = config.Channels.Signal.DmPolicy;

                            if (policy is "closed")
                                continue; // Silently drop all inbound messages

                            if (!msg.IsSystem && policy is "pairing" && !pairingManager.IsApproved(msg.ChannelId, msg.SenderId))
                            {

Copilot · 2026-04-01T05:32:28Z

src/OpenClaw.Gateway/Extensions/GatewayWorkers.cs

+                            if (config.Routing.Enabled)
+                            {
+                                var routeResolver = new OpenClaw.Gateway.Integrations.AgentRouteResolver(config.Routing);


AgentRouteResolver is instantiated per inbound message. Since it only depends on config, this adds avoidable allocations on a hot path. Consider constructing it once (outside the loop) or registering/injecting a singleton resolver and reusing it for all messages.

Suggested change

if (config.Routing.Enabled)

{

var routeResolver = new OpenClaw.Gateway.Integrations.AgentRouteResolver(config.Routing);

static OpenClaw.Gateway.Integrations.AgentRouteResolver? routeResolver;

if (config.Routing.Enabled)

{

routeResolver ??= new OpenClaw.Gateway.Integrations.AgentRouteResolver(config.Routing);

Copilot · 2026-04-01T05:32:29Z

src/OpenClaw.Gateway/Extensions/GatewayWorkers.cs

+                                // Append verbose mode footer (tool calls and token delta)
+                                if (session.VerboseMode)
+                                {
+                                    var lastTurn = session.History.Count > 0 ? session.History[^1] : null;
+                                    var toolCallCount = lastTurn?.ToolCalls?.Count ?? 0;
+                                    responseText += $"\n\n---\n🔧 {toolCallCount} tool call(s) | ↑ {inputTokenDelta} in / {outputTokenDelta} out tokens (this turn)";
+                                }


The verbose footer’s tool call count is likely incorrect: RunAsync appends the final assistant text as the last history turn (without ToolCalls), while tool invocations are stored on earlier [tool_use] turns. As a result, session.History[^1].ToolCalls will typically be null and the footer will report 0 even when tools were used. Track tool calls for the current turn explicitly (e.g., capture history/tool-call counts before the run and diff after, or expose per-turn tool-call count from the runtime).

Copilot · 2026-04-01T05:32:29Z

src/OpenClaw.Gateway/Composition/RuntimeInitializationExtensions.cs

+            {
+                var before = session.History.Count;
+                await concreteRuntime.CompactHistoryAsync(session, ct);
+                return Math.Min(config.Memory.CompactionKeepRecent, session.History.Count);


The /compact callback returns Math.Min(config.Memory.CompactionKeepRecent, session.History.Count) after compaction, but session.History.Count includes the inserted summary turn, so this can over-report “recent turns preserved” by 1 when compaction runs (and more generally won’t match the runtime’s keepCount). Also, before is assigned but never used. Compute and return the actual preserved turn count (e.g., based on the pre-compaction history size and keep-recent rules), and remove the unused local.

Suggested change

return Math.Min(config.Memory.CompactionKeepRecent, session.History.Count);

var preserved = Math.Min(config.Memory.CompactionKeepRecent, before);

return preserved;

Copilot · 2026-04-01T05:32:29Z

src/OpenClaw.Agent/AgentRuntime.cs

    /// <summary>
    /// Compacts session history by summarizing older turns via the LLM.
    /// Keeps the most recent turns verbatim and replaces older ones with a summary.
    /// </summary>
-    internal async Task CompactHistoryAsync(Session session, CancellationToken ct)
+    public async Task CompactHistoryAsync(Session session, CancellationToken ct)
    {
        if (session.History.Count <= _compactionThreshold)


CompactHistoryAsync was made public so the gateway can call it, but it’s still not exposed via IAgentRuntime, forcing callers to downcast (agentRuntime is AgentRuntime). This couples the gateway to a concrete orchestrator and makes it easy to miss compaction support in other runtimes. Consider introducing an interface (e.g., IHistoryCompactor) or adding a capability method on IAgentRuntime so the gateway can wire /compact without concrete casts.

…tegrations Complete README overhaul covering the current feature set: - 48 native tools organized by category with full table - 9 channel adapters with transport details and feature highlights - Tool presets (full/coding/messaging/minimal) and groups (group:*) - Chat commands (/think, /compact, /verbose) - Plugin installer (openclaw plugins install) - Multi-agent routing, Tailscale, Gmail Pub/Sub, mDNS - Updated architecture diagram showing message pipeline, session manager, route resolution, and tool backends - Simplified runtime flow diagram - Badges for tool count and channel count Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Move AgentRouteResolver construction outside worker loop (was per-message) - Route model override now always applies (was ??= which skipped existing) - Fix verbose footer tool call count: scan all turns since last user turn instead of only checking the final assistant turn - Add verbose footer to streaming branch (was non-streaming only) - Fix /compact callback: return actual remaining turn count, remove unused var - Wire reasoning_effort into MafAgentRuntime.CreateChatOptions for MAF parity Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Copilot AI review requested due to automatic review settings April 1, 2026 05:27

Copilot started reviewing on behalf of Telli April 1, 2026 05:27 View session

augmentcode bot reviewed Apr 1, 2026

View reviewed changes

Copilot AI reviewed Apr 1, 2026

View reviewed changes

Telli and others added 2 commits March 31, 2026 22:35

Telli merged commit 14f4072 into feat/channel-expansion-tools-features Apr 1, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: wire deferred review items end-to-end#39

fix: wire deferred review items end-to-end#39
Telli merged 3 commits intofeat/channel-expansion-tools-featuresfrom
fix/review-followup-wiring

Telli commented Apr 1, 2026

Uh oh!

augmentcode bot commented Apr 1, 2026

Uh oh!

augmentcode bot left a comment

Uh oh!

augmentcode bot Apr 1, 2026

Uh oh!

augmentcode bot Apr 1, 2026

Uh oh!

augmentcode bot Apr 1, 2026

Uh oh!

augmentcode bot Apr 1, 2026

Uh oh!

augmentcode bot Apr 1, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	return Math.Min(config.Memory.CompactionKeepRecent, session.History.Count);
	var preserved = Math.Min(config.Memory.CompactionKeepRecent, before);
	return preserved;

Conversation

Telli commented Apr 1, 2026

Summary

Test plan

Uh oh!

augmentcode bot commented Apr 1, 2026

Uh oh!

augmentcode bot left a comment

Choose a reason for hiding this comment

Uh oh!

augmentcode bot Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

augmentcode bot Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

augmentcode bot Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

augmentcode bot Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

augmentcode bot Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants