Stop letting AI agents
drift. Execute.

Structured task execution for AI agents. No lost context, no skipped steps. If your agent runs out of tokens, just swap in a new one and keep going.

$ curl -fsSL https://raw.githubusercontent.com/superplan-md/superplan-plugin/main/scripts/install.sh | sh
PS> curl.exe -fsSL -o install-superplan.cmd https://raw.githubusercontent.com/superplan-md/superplan-plugin/main/scripts/install.cmd; if ($LASTEXITCODE -eq 0) { .\install-superplan.cmd }
> curl.exe -fsSL -o install-superplan.cmd https://raw.githubusercontent.com/superplan-md/superplan-plugin/main/scripts/install.cmd && install-superplan.cmd
Then run
superplan init
in your repo.
The CLI is designed for agents to follow, not for humans to run manually. You define the work; your agent executes it through Superplan's structured runtime.
54GitHub Stars
25Releases
Apache 2.0License

What it does

Forced step order
Tasks run in sequence. Agents and humans cannot skip ahead. Every step is a contract, not a suggestion.
Durable context
All state lives in .superplan/ inside your repo. No external service, no sync issues, survives reboots and agent swaps.
Pause and resume
Token limit hit? Switched models? Drop back in, run superplan run, and continue from the exact step where things stopped.
JSON-first output
Every command outputs structured JSON. Pipe it into agents, scripts, or dashboards without parsing hacks.
Skills system
Built-in execution disciplines for planning, debugging, validation, and review. Running behind the scenes on every task.
Agent-controlled loop
Designed for AI agents to drive, not humans to babysit. The CLI is the control plane your agent talks to.

How it works

01

Define a change

Every piece of work starts with a named change. The scope is explicit from the start, no ambiguity.

$ superplan change new my-feature --json
02

Tasks are created and tracked

Superplan converts your change into ordered, trackable tasks. Each task has a contract with no vague intentions.

$ superplan change task add my-feature --title "..."
03

Agents execute through the CLI

Your AI agent runs through tasks one by one. It can't skip steps. Progress is saved after every action.

$ superplan run --json
04

Status is always visible

Check what's done, what's next, and what's blocked at any time. If your agent hits a token limit or you switch to a different model entirely, state is safe in the repo. Any agent can resume from the exact same point.

$ superplan status --json

What people are saying

"First time I've had an AI agent actually finish what it started. Context survives across sessions, steps can't be skipped. It just works."

@devops_nico

"My Claude session hit its token limit mid-refactor. I switched to GPT-4o, ran superplan run, and it picked up at the exact next task. Genuinely shocked."

@buildingwithbri

"This is what I wanted project management tools to be for agents. Finally something that forces order instead of hoping the agent remembers."

@kartik_dev

"The JSON-first design is underrated. I pipe superplan status into a dashboard that shows my whole team what the agent is doing."

@liz_eng

"Been using it for two weeks. The thing agents kept failing at was losing the plan mid-conversation. Superplan just removes that failure mode."

@tomhammers

"The .superplan/ folder living in my repo means the whole team can see what the agent did and what's next. No black box."

@rachael_builds

"Ran a 3-hour coding session across two different models. Each knew exactly where to pick up. That's the killer feature."

@srdjanp

"Simple concept, surprisingly deep. The skills system runs discipline around every task automatically. Planning, debugging, review — it's all there."

@mira_infra

"Feels like what GitHub Issues should have been for AI-driven work. Tasks with contracts, not just sticky notes."

@andrewmakes

"I was skeptical about yet another CLI tool. Three weeks in, superplan run is the first command I run in every session."

@elise_codes

"The comparison vs chat-based work is night and day. No more re-explaining context. The agent reads the task, does it, moves on."

@productdev

"Open source, lives in your repo, no external service. This is what agent tooling should look like."

@jasper_oss

Superplan vs everything else

How structured CLI execution compares to the alternatives teams rely on today.

Feature Superplan Chat-based AI TODO Lists Project Managers
Persistent context Full history Lost per session ~ Manual notes ~ Docs only
Agent-native CLI-first ~ Chat only Not designed for agents Human-only UX
Enforced step order Blocked gates Improvises freely No enforcement ~ Manual ordering
Lives in your repo .superplan/ External service External app External service
Swap agents mid-task Built-in Start from scratch No state ~ With setup
Zero drift guarantee Enforced Prone to drift Untracked Manual reviews
Open source Apache 2.0 Proprietary ~ Varies Often proprietary

Up and running in 60 seconds

Terminal
$ curl -fsSL https://raw.githubusercontent.com/superplan-md/superplan-plugin/main/scripts/install.sh | sh
Installing superplan...
✓ superplan installed successfully
 
$ superplan init
✓ Superplan initialized in /your-repo/.superplan
 
$ superplan change new my-feature --json
"change_id": "my-feature",
"root": ".superplan/changes/my-feature"
 
$ superplan run --json
✓ Executing T-001: Set up project scaffold

Ready to stop drifting?

Install Superplan in your repo and start executing with structure. Takes less than a minute.