Slack Advanced Actions Bot (SAB)

A smart Slack bot powered by LangGraph and local Qwen3-8B (with optional Gemini fallback) for agentic workflow orchestration.

Features

Feature	Command	Description
Summarize	`/sab summarize` or mention	Summarizes recent messages using local LLM
Reminders	`/sab -r "task" @30m`	Set timed reminders (persisted in SQLite)
Natural Language Reminders	`/sab remind me to...`	dateparser handles "tomorrow at 3pm"
List Reminders	`/sab reminders`	Show all pending reminders with IDs
Cancel Reminder	`/sab reminder cancel <id>`	Cancel by job ID
GitHub Issue/PR Lookup	`owner/repo#123` or paste URL	Fetches issue/PR details
Latest Issues	`/sab latest issues`	Newest open issues across all repos
Latest PRs	`/sab latest prs`	Newest open PRs across all repos
Code Review	`/sab codereview owner/repo#123`	3-subagent review (Security + Performance + Best Practices)
PR Risk Score	(auto in code review)	🔴 High / 🟡 Medium / 🟢 Low from Semgrep + LLM
Semgrep Security	(auto in code review)	Real static analysis grounding
Learning Paths	`/sab learn <topic>`	3-agent research → structure → resources
Web Search	(auto in /learn)	Tavily API for real URLs
MCP Source Footer	(auto in review/learn)	Visible "via GitHub MCP" or fallback note
Slack mrkdwn Auto-fix	(auto)	Bold, links, and bare URLs converted to Slack format
Help	`/sab` or "what can you do"	Static command list
System Diagnostics	`/sab test`	Checks LLM provider + all 3 MCP sessions
Daily Digest	`/sab digest subscribe`	Proactive daily GitHub issues/PRs post
Digest Demo	`/sab digest demo`	Preview digest in ~2 minutes
Duplicate Check	`/sab duplicate owner/repo "title"`	Find similar open issues via difflib
Release Notes	`/sab release notes owner/repo`	Generate notes from merged PRs
Dual LLM	`LLM_PROVIDER=local\|gemini`	Local Qwen3 primary, Gemini cross-fallback
Typo Tolerance	"coderview", "review", "pr"	difflib fuzzy matching
Real-Time Search	`/sab search <query>` or `@bot find discussions about X`	Cross-channel search via Slack RTS API

Architecture

Built with LangGraph StateGraph for deterministic, debuggable agentic workflows:

┌─────────────────────────────────────────────────────────────────────┐
│                        Slack Workspace                              │
└────────────────────────────┬────────────────────────────────────────┘
                             │ Socket Mode (WebSocket)
                             ▼
┌─────────────────────────────────────────────────────────────────────┐
│  main.py (Bolt App)                                                 │
│  ├── /sab command ──► cmd_sab() ──► handle_sab_command()           │
│  ├── app_mention ──► on_mention() ──► handle_app_mention()         │
│  └── message ──► on_message() ──► handle_message_event()           │
│                                                                      │
│  [trigger_id dedup on /sab] [event_ts dedup on mention/message]    │
│  [processing reaction decorator] [MCP init in background thread]   │
└────────────────────────────┬────────────────────────────────────────┘
                             ▼
┌─────────────────────────────────────────────────────────────────────┐
│  graph/workflow.py (LangGraph StateGraph)                           │
│  ┌──────────┐                                                       │
│  │ classify │ ◄── entry_point                                       │
│  └────┬─────┘                                                       │
│       │ 15 conditional routes                                       │
│       ├──► reminder (parse → schedule / list / cancel)              │
│       ├──► github (extract → fetch → response)                      │
│       ├──► context (summarize → response)                           │
│       ├──► learn (research → structure → resources → response)      │
│       ├──► codereview (fetch → [security,performance,best] → merge) │
│       ├──► search (RTS via assistant.search.context)                │
│       ├──► digest / duplicate / release_notes                       │
│       ├──► help / greeting / chat / test_llm                        │
└────────────────────────────┬────────────────────────────────────────┘
                             ▼
┌─────────────────────────────────────────────────────────────────────┐
│  services/                                                          │
│  ├── llm_service.py ──► llama-server (Qwen3-8B, localhost:8080)    │
│  ├── github_service.py ──► GitHub API (TTL cache, rate-limit track) │
│  ├── codereview_service.py ──► Semgrep + 3 LLM subagents           │
│  ├── learn_service.py ──► Tavily search + GitHub repos + LLM        │
│  ├── reminder_service.py ──► SQLite-backed APScheduler              │
│  ├── slack_search_service.py ──► Slack RTS (assistant.search)       │
│  └── mcp_client.py ──► GitHub/Fetch/Slack MCP servers               │
└─────────────────────────────────────────────────────────────────────┘

Public Repo Access

GitHub lookups work for public repos without a token. Set GITHUB_TOKEN in .env for private repos and higher rate limits (5,000/hr vs 60/hr unauthenticated).

Tech Stack

LangGraph — StateGraph workflow orchestration with fan-out/fan-in
Slack Bolt — Socket Mode event handling
Local Qwen3-8B — LLM via llama-server (zero API cost, no data leaves machine)
APScheduler — SQLite-persisted reminder scheduling
GitHub API — Issue/PR lookup with TTL cache (works without token for public repos)
MCP — Model Context Protocol for extensible tool access (GitHub, Fetch, custom Slack server)
Semgrep — Real static analysis for security reviews (pinned ruleset, no network lookup)
Tavily — Web search for verified learning resources
dateparser — Natural language time parsing
difflib — Typo-tolerant command matching
Pydantic — Type-safe state models

Project Structure

Slack-Action-Bot/
├── main.py                      # Bolt app, MCP init, dedup guards (trigger_id + event_ts)
├── config.py                    # Env vars (tokens, endpoints)
├── graph/
│   ├── state.py                 # BotState TypedDict, ReminderData
│   ├── nodes.py                 # 30+ node functions
│   └── workflow.py              # StateGraph build, routing logic
├── handlers/
│   ├── commands.py              # /sab slash command handler
│   ├── events.py                # app_mention + message handlers
│   └── shared.py                # is_real_message, fetch_*, build_initial_state
├── services/
│   ├── llm_service.py           # Local LLM calls, PERSONA, summarize
│   ├── github_service.py        # GitHub API, TTL cache, rate-limit tracking
│   ├── codereview_service.py    # 3-subagent review, Semgrep, risk score
│   ├── learn_service.py         # 3-agent learning path, Tavily search
│   ├── reminder_service.py      # SQLite-backed APScheduler
│   ├── mcp_client.py            # MCP AsyncExitStack, background loop, _evict_session()
│   ├── mcp_slack_server.py      # Custom Slack MCP server (stdio)
│   ├── slack_summarize_service.py # MCP-first channel summary
│   ├── slack_search_service.py   # Real-time search via assistant.search.context
│   └── slack_features.py        # Processing reaction decorator
├── reminders.db                 # SQLite persistent reminders
├── test_all.py                  # 168 unit tests
├── test_e2e.py                  # 57 end-to-end tests
└── pyproject.toml               # Dependencies (uv-managed)

Setup

Start llama-server with Qwen3-8B:

cd ~/llama.cpp
./build/bin/llama-server \
  -m <path-to>/Qwen3-8B-Q4_K_M.gguf \
  -ngl 999 \
  -c 16384 \
  --parallel 4 \
  --host 0.0.0.0 \
  --port 8080

Copy .env.example to .env and fill in:

SLACK_BOT_TOKEN=xoxb-...
SLACK_APP_TOKEN=xapp-...
SLACK_SIGNING_SECRET=...
GITHUB_TOKEN=ghp-...           # optional for public repos
LLAMA_BASE_URL=http://localhost:8080
LLAMA_PARALLEL=4
LLM_PROVIDER=local
LLM_FALLBACK_ENABLED=true
GOOGLE_API_KEY=...              # optional, Gemini fallback when local fails
GEMINI_MODEL=gemini-2.0-flash
TAVILY_API_KEY=tvly-...         # optional, for /learn web search

Install dependencies:

uv sync
uv sync --group optional   # installs semgrep for code review static analysis

Enable Real-Time Search (optional):
- Go to https://api.slack.com/apps → your app → App Home → enable Assistant
- Add Bot Token Scopes: search:read.public, search:read.files, search:read.users
- Reinstall app to workspace
Run the bot:
```
uv run main.py
```

Commands

Command	Description
`/sab`	Show help / command list
`/sab summarize`	Summarize channel or thread messages
`/sab -r "task" @30m`	Set a reminder in 30 minutes
`/sab remind me to...`	Natural language reminder
`/sab reminders`	List pending reminders
`/sab reminder cancel <id>`	Cancel a reminder
`/sab latest issues`	Newest open issues across all repos
`/sab latest prs`	Newest open PRs across all repos
`/sab codereview owner/repo#123`	3-perspective code review with risk score
`/sab learn <topic>`	Structured learning path
`/sab digest subscribe`	Daily GitHub digest at 9am UTC
`/sab digest demo`	Preview digest in ~2 minutes
`/sab duplicate owner/repo "title"`	Find similar open issues
`/sab release notes owner/repo`	Generate release notes from merged PRs
`/sab test`	Diagnostics: LLM provider + MCP health check
`/sab search <query>`	Real-time cross-channel search (needs `search:read` scopes)
`@bot find discussions about X`	Same, via @mention (action_token required)
`owner/repo#123`	Auto-fetch issue/PR details
Paste GitHub URL	Auto-fetch issue/PR details

Name		Name	Last commit message	Last commit date
.Old		.Old
graph		graph
handlers		handlers
services		services
.env.example		.env.example
.gitignore		.gitignore
.python-version		.python-version
CHANGELOG.md		CHANGELOG.md
CHANGES.md		CHANGES.md
DESIGN.md		DESIGN.md
Diagram.svg		Diagram.svg
Flowchat.webp		Flowchat.webp
LICENSE		LICENSE
README.md		README.md
config.py		config.py
documentration.md.txt		documentration.md.txt
env.example		env.example
main.py		main.py
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt
test_all.py		test_all.py
test_e2e.py		test_e2e.py
uv.lock		uv.lock

Navigation Menu

Repository navigation

About

Repository files navigation

Slack Advanced Actions Bot (SAB)

Features

Architecture

Public Repo Access

Tech Stack

Project Structure

Setup

Commands

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

Slack Advanced Actions Bot (SAB)

Features

Architecture

Public Repo Access

Tech Stack

Project Structure

Setup

Commands

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages