═══════════════════════════════════════════════════════════════════════════════════════
█▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀█
█ █
█ ██╗░░░░░░█████╗░██████╗░░██████╗████████╗███████╗██████╗░ █
█ ██║░░░░░██╔══██╗██╔══██╗██╔════╝╚══██╔══╝██╔════╝██╔══██╗ █
█ ██║░░░░░██║░░██║██████╦╝╚█████╗░░░░██║░░░█████╗░░██████╔╝ █
█ ██║░░░░░██║░░██║██╔══██╗░╚═══██╗░░░██║░░░██╔══╝░░██╔══██╗ █
█ ███████╗╚█████╔╝██████╦╝██████╔╝░░░██║░░░███████╗██║░░██║ █
█ ╚══════╝░╚════╝░╚═════╝░╚═════╝░░░░╚═╝░░░╚══════╝╚═╝░░╚═╝ █
█ █
█ ██╗░░░░░░░██╗░█████╗░██████╗░██╗░░██╗███████╗██╗░░░░░░█████╗░░██╗░░░░░░░██╗░██████╗ █
█ ██║░░██╗░░██║██╔══██╗██╔══██╗██║░██╔╝██╔════╝██║░░░░░██╔══██╗░██║░░██╗░░██║██╔════╝ █
█ ╚██╗████╗██╔╝██║░░██║██████╔╝█████═╝░█████╗░░██║░░░░░██║░░██║░╚██╗████╗██╔╝╚█████╗░ █
█ ░████╔═████║░██║░░██║██╔══██╗██╔═██╗░██╔══╝░░██║░░░░░██║░░██║░░████╔═████║░░╚═══██╗ █
█ ░╚██╔╝░╚██╔╝░╚█████╔╝██║░░██║██║░╚██╗██║░░░░░███████╗╚█████╔╝░░╚██╔╝░╚██╔╝██████╔╝ █
█ ░░╚═╝░░░╚═╝░░░╚════╝░╚═╝░░╚═╝╚═╝░░╚═╝╚═╝░░░░░╚══════╝░╚════╝░░░╚═╝░░░╚═╝░╚═════╝░ █
█ █
█▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄█
═════════════════════════════════════════════════════════════════════════════════════════
🦞 FEEL THE FLOW OF THE LOBSTER 🦞
PR REVIEW ● DEPLOY ANNOUNCE ● ROLL CALL ● X POST
● KNOWLEDGE EXTRACTION ● AND MORE...
══════════════════════════════════════════════════════════════════════
DETERMINISTIC PIPELINES ● APPROVAL GATES ● STATE TRACKING
══════════════════════════════════════════════════════════════════════
©2026 JDRHYNE ENTERTAINMENT
A collection of ready-to-use Lobster workflow templates for Moltbot (formerly Clawdbot).
Lobster is a deterministic workflow engine — typed pipelines with approval gates, state tracking, and token-efficient automation. These workflows replace freeform LLM orchestration with predictable, repeatable pipelines.
Think of them like shell scripts for AI agents: instead of hoping the LLM figures out the right sequence every time, you define the exact steps, checkpoints, and approvals once — and it runs the same way every time.
Looking for agent skills and prompts instead of workflows? Check out agent-skills — a companion repo of skills for Moltbot, Claude Code, Codex, and other LLM agents.
🤝 Want to contribute a workflow? See Contributing below — PRs welcome!
| Workflow | Steps | Description |
|---|---|---|
x-morning-post |
8 | Select a content seed, generate a draft via LLM, approve, and post to X |
daily-roll-call |
10 | Create a thread, ping all agents, collect reports, compile a fleet summary |
pr-review |
7 | Fetch GitHub PR state, diff against last run, summarize changes, post update |
deploy-announce |
8 | Fetch release info, generate announcement, approve, post to channel + X |
knowledge-extraction |
13 | Read daily notes, extract durable facts, write to a knowledge graph |
| Without Lobster | With Lobster |
|---|---|
| LLM re-plans every step | Deterministic pipeline |
| 10+ tool calls per run | 1 workflow call |
| "Please don't send" (hope) | approve = hard stop |
| Forgets what it did yesterday | Stateful — tracks cursors |
Lobster ships with Moltbot. If you have Moltbot, you have Lobster.
# Clone this repo
git clone https://github.com/jdrhyne/lobster-workflows.git
# Copy a workflow to your Moltbot workspace
cp lobster-workflows/workflows/pr-review.lobster ~/my-workspace/workflows/Each workflow has a Platform Customization section at the top explaining what to adapt:
# ── Platform Customization ──────────────────────────────────────────────
# This workflow uses Moltbot's message tool for notifications.
# Adapt these for your messaging platform:
# - channel IDs → your platform's channel/group/chat identifiers
# - thread-create → may not be available on all platforms
# - message formatting → some platforms use markdown, others don't
# ────────────────────────────────────────────────────────────────────────All channel/bot references use placeholders like YOUR_CHANNEL_ID and BOT_ID_1 — replace these with your actual values.
# Run a workflow
lobster run workflows/pr-review.lobster --args-json '{"repo":"owner/repo","pr":"123"}'
# Run with custom inputs
lobster run workflows/deploy-announce.lobster --args-json '{"repo":"owner/repo","channel":"123456"}'
# Run knowledge extraction (no messaging required)
lobster run workflows/knowledge-extraction.lobsterHuman-in-the-loop content publishing pipeline.
Parse content seeds → Select seed → Confirm → Generate draft (LLM) → Approve → Post via bird CLI → Mark posted
Inputs: seed_id (optional), seeds_file (path to your content seeds)
Tools: bird CLI, LLM via clawd.invoke
Approval gates: 2 (seed selection + final post)
Multi-agent coordination pipeline. Collects status reports from all agents and compiles a summary.
Create thread → Ping agents → Request reports → Self-report → Wait → Read responses → Compile summary → Post
Inputs: reports_channel, wait_minutes, bot_ids, agent_labels
Tools: Moltbot message + sessions_send
Approval gates: 0 (fully automated)
Note: Requires a platform with threading support (Discord, Slack). For non-threaded platforms, use a dedicated channel.
GitHub PR monitoring with change detection. Only reports when something changes.
Fetch PR details → Check CI status → Diff against last run → Generate summary (LLM) → Approve → Post → Save state
Inputs: repo (required), pr (required), channel (optional), changes_only (default: true)
Tools: gh CLI, LLM via clawd.invoke
Approval gates: 1 (before posting)
Release announcement pipeline with optional cross-posting.
Fetch release → Find previous tag → Fetch changelog → Generate announcement (LLM) → Approve → Post to channel → Optionally post to X → Save state
Inputs: repo (required), channel (required), tag (optional, defaults to latest), post_to_x (default: false)
Tools: gh CLI, bird CLI, LLM via clawd.invoke
Approval gates: 2 (channel post + X post)
Extract durable facts from daily notes into a structured knowledge graph. Pairs with the knowledge-graph skill.
Read daily notes → List existing entities → Extract facts (LLM) → Review → Approve → Create entities → Write facts → Log → Save state
Inputs: date (default: today), auto_approve (default: false), workspace (default: current dir)
Tools: LLM via clawd.invoke, filesystem
Approval gates: 1 (fact approval, can be auto-skipped)
No messaging required — this workflow is entirely file-based.
These workflows use Moltbot's message tool, which routes to whatever platform you have configured:
| Platform | Threads | Notes |
|---|---|---|
| Discord | ✅ | Full support — threads, reactions, embeds |
| Slack | ✅ | Full support — threads, reactions |
| Telegram | ❌ | Use dedicated channels instead of threads |
| Signal | ❌ | Use dedicated groups instead of threads |
Workflows that don't use messaging (like knowledge-extraction) work on any platform or none at all.
name: my-workflow
description: What it does
version: "1.0.0"
inputs:
my_input:
type: string
description: What this input controls
default: "sensible-default"
steps:
- id: step_1
description: What this step does
command: shell-command-here
output: json
- id: step_2
description: LLM-powered step
command: clawd.invoke
args:
tool: sessions_spawn
task: "Generate something based on ${step_1.stdout}"
- id: approval
description: Human checkpoint
command: approve
args:
prompt: "Ready to proceed?"
approval: required
- id: final_step
description: Do the thing
command: some-command
condition: $approval.approved
output:
success: $final_step.successHave a useful workflow? PRs welcome!
- Add your
.lobsterfile toworkflows/ - Include a Platform Customization comment block
- Use placeholder values (
YOUR_CHANNEL_ID, etc.) — not real IDs - Add a description to this README
- Submit a PR
- Lobster — The workflow engine
- Moltbot — The AI agent platform
- Agent Skills — Skills and prompts for AI agents
MIT
Jonathan Rhyne (@jdrhyne)