feat: add Slack bot integration for kagent #1017

apexlnc · 2025-10-13T21:28:02Z

Summary

Production-ready Slack bot integration for the Kagent multi-agent platform. Provides a unified Slack interface to interact with multiple AI agents through DMs and channel mentions.

Features

Core Functionality

Socket Mode Integration: WebSocket-based connection (no public endpoint required)
Dynamic Agent Discovery: Automatically discovers agents from /api/agents with 5-minute caching
Intelligent Routing: Keyword-based matching routes messages to appropriate agents
Thread Context Isolation: Each thread maintains independent conversation context
Session Management: Full conversation history maintained across multiple turns

User Experience

Streaming Responses: Real-time updates with progressive message rendering (2s intervals)
Human-in-the-Loop Approval: Smart detection of approval requests with ✅/❌ buttons
Rich Formatting: Slack Block Kit responses with agent metadata
Slash Commands: /agents to list, /agent-switch to manually select agents

Security & RBAC

Slack User Group Integration: Check group membership with 5-minute caching
Agent-Level Permissions: Configurable YAML-based access control
Email Allowlists: Support for individual user permissions
Configurable Deny Messages: Custom access denial messages per agent

Architecture

handlers/: Slack event handlers (mentions, commands, actions, middleware)
services/: Business logic (A2A client, agent discovery, routing)
auth/: RBAC with Slack group membership checking
slack/: Message formatting and validation utilities

Testing

Tested locally with kagent deployment using Socket Mode in DM-only configuration.

apexlnc · 2025-10-13T21:32:36Z

cc @peterj @EItanya @yuval-k @ilackarms can I get a review on this?

FWIW, i've fully tested this on a Slack enterprise-grid account and this is loosely based off existing patterns I'm currently using for other use cases.

The HITL portion is very "basic" atm with regex matching, but is a good starter for using Slack as an approval gate down the road -- using fancier conditions.

There are a few minor things that I'd like to wire up, but wanted to make sure this is line with the direction the project is heading towards.

apexlnc · 2025-10-16T11:47:31Z

@EItanya can I get a review on this? Not sure which maintainer I should be tagging

peterj · 2025-10-20T11:57:55Z

slackbot/Makefile

+	@echo "  make test        - Run tests (if available)"
+	@echo "  make lint        - Run linting with ruff"
+	@echo "  make type-check  - Run type checking with mypy"
+	@echo "  make build       - Build Docker image"


make build doesn't work for me:

Should there be a default config folder with some default permissions?

Yes, defer to the wider audience on sensible defaults.

slackbot/SLACK_SETUP.md

peterj · 2025-10-20T12:11:19Z

This is a great feature - I got it to work; however, I think this Slack bot should probably be in a separate repo, instead of having it in the main repository. @EItanya what do you think?

Also, it would be great starting point (in terms of UX and functionality) for creating a Discord and/or MS Teams bots as well.

peterj · 2025-10-20T12:13:17Z

@apexlnc how do I continue the conversation with the agent? I see that it creates a thread when using @kagent-bot, but if I need to continue the same conversation, I have to do another @kagent-bot within the thread it seems? I wonder if it would make sense to automatically chat with the agent when in thread (e.g. without prefixing each message with kagent-bot) - what do you think?

peterj

added a couple of comments

slackbot/src/kagent_slackbot/handlers/mentions.py

slackbot/src/kagent_slackbot/handlers/middleware.py

slackbot/src/kagent_slackbot/main.py

slackbot/src/kagent_slackbot/services/a2a_client.py

slackbot/src/kagent_slackbot/services/agent_discovery.py

slackbot/src/kagent_slackbot/slack/formatters.py

slackbot/src/kagent_slackbot/handlers/middleware.py

apexlnc · 2025-10-20T14:44:22Z

@apexlnc how do I continue the conversation with the agent? I see that it creates a thread when using @kagent-bot, but if I need to continue the same conversation, I have to do another @kagent-bot within the thread it seems? I wonder if it would make sense to automatically chat with the agent when in thread (e.g. without prefixing each message with kagent-bot) - what do you think?

You should be able to continue the conversation in the thread. Are you not able to?

This is a great feature - I got it to work; however, I think this Slack bot should probably be in a separate repo, instead of having it in the main repository. @EItanya what do you think?

Also, it would be great starting point (in terms of UX and functionality) for creating a Discord and/or MS Teams bots as well.

Sure, let me know if you want it in another repo, and if so, which repo.

@peterj -- btw, a lot of the "magic" happens once this is merged : #1025

i'm also realizing that i didn't push some of my local changes (mostly housekeeping stuff) -- will do that later after some responses to the comments above

Adds a production-ready Slack bot that provides a unified interface to interact with Kagent agents through Slack, including full Human-in-the-Loop (HITL) support for tool approvals. Key features: - Socket Mode bot for @mentions and direct messages - Dynamic agent discovery and intelligent routing with keyword-based routing - Real-time streaming responses with HITL approval workflow - RBAC with Slack user group integration - Prometheus metrics with kagent_slackbot namespace prefix - Complete Kubernetes manifests (Deployment, HPA, PDB, RBAC) - Dockerfile for containerized deployment Architecture: - Uses A2A SDK types throughout for type safety and validation - Pydantic models for agent metadata and HITL protocol - Handles both TaskStatusUpdateEvent and TaskArtifactUpdateEvent - Structured DataPart decisions for tool approval workflow - Interactive approval buttons with streaming completion responses The bot integrates with Kagent's A2A protocol via JSON-RPC 2.0 and supports all agent types (Declarative, BYO, etc.). Follows bolt-python async patterns. Addresses maintainer feedback: - Use Pydantic models for agent data validation - Full A2A SDK type integration (not raw dicts) - Metrics namespace prefix for Prometheus - Code formatted with ruff - Default permissions.yaml config provided Signed-off-by: apexlnc <[email protected]>

marcinkubica · 2025-10-21T08:36:19Z

@apexlnc how do I continue the conversation with the agent? I see that it creates a thread when using @kagent-bot, but if I need to continue the same conversation, I have to do another @kagent-bot within the thread it seems? I wonder if it would make sense to automatically chat with the agent when in thread (e.g. without prefixing each message with kagent-bot) - what do you think? Screenshot 2025-10-20 at 2 11 32 PM

You should be able to continue the conversation in the thread. Are you not able to?

@peterj This is probably for the better: bot should only respond to a configurable calls (ie: @kagent kagent) and not by default to any message in the thread created.

@apexlnc I'm noticing also the emoji response 👀 are not removed after bot responded - they should be present when bot is processing and removed after bot responded.

Also, something nice to steal from Onyx (in the later time) - "continue in kagent" and "I need help from a human"

…plication Address reviewer feedback from PR review: - Run ruff formatter on all Python files (all checks pass) - Remove 👀 emoji reaction after bot response completion - Fix Docker build by including config/ directory in image - Reorganize SLACK_SETUP.md to clearly separate Option A (manifest) from Option B (manual setup) Additional code quality improvements: - Extract duplicate button parsing logic to _parse_button_value() helper - Extract duplicate SSE streaming logic to _parse_sse_stream() helper - Replace magic numbers with named constants: * SLACK_TEXT_SUMMARY_LENGTH = 200 * PREVIEW_MAX_LENGTH = 1000 * UPDATE_THROTTLE_SECONDS = 2 - Use existing AGENT_CACHE_TTL constant in main.py - Add emoji removal to all exit paths including errors and HITL flows - Add _remove_reaction() helpers to eliminate further duplication This eliminates ~60 lines of duplicate code and improves maintainability without changing any functionality. Fixes: peterj's review comments on make build, docs, formatting Signed-off-by: apexlnc <[email protected]>

apexlnc · 2025-10-26T23:58:41Z

Thanks for the feedback @peterj @marcinkubica

I believe all the comments have been addressed, but let me know if I missed anything!

apexlnc mentioned this pull request Oct 13, 2025

feat: add Slack bot integration for kagent apexlnc/kagent#1

Closed

apexlnc force-pushed the feat/slack-bot-integration branch from cd76e9f to 50c8b0a Compare October 13, 2025 21:28

apexlnc force-pushed the feat/slack-bot-integration branch 5 times, most recently from 8789d2e to fd2d66f Compare October 14, 2025 03:06

apexlnc force-pushed the feat/slack-bot-integration branch from fd2d66f to aafe032 Compare October 17, 2025 02:56

apexlnc requested review from EItanya, ilackarms and yuval-k as code owners October 17, 2025 02:56

apexlnc force-pushed the feat/slack-bot-integration branch 4 times, most recently from adab011 to f67f765 Compare October 18, 2025 21:24