Stable on PyPI

Give your AI agent a plan, not a prompt.

Sembl reads your repo and turns a rough task into a precise Work Order - what to change, what to leave alone, and how to prove it worked. Your agent executes inside the lines instead of guessing.

Star on GitHub

sembl - powershell

            $ pip install sembl
            Successfully installed sembl 0.1.9
             
            $ sembl generate --task "fix the failing login redirect test"
              reading repo ... graph context: on
              Work Order ready
              scope: 2 files editable, 3 read-only
              proof: 4 validation commands, risk: medium
          

Locks per Work Order

LLM providers

Output files

Command to run

How it works

From a one-line task to a plan an agent can follow.

No new workflow to learn. You describe the task, Sembl handles the scoping, and your existing agent does the work. The Work Order is the stable layer; the agent underneath is swappable.

Describe the task

Plain words, however rough. "Fix the login redirect test" is enough to start - no schema, no ceremony.

Sembl reads the repo

It probes structure and branch state, and when graph tools are available, pulls deeper context to find what's actually involved.

Graphifycode-review-graph

It writes the Work Order

Goal, editable paths, read-only zones, acceptance criteria, and stop conditions - grounded in your real files, not invented ones.

Your agent executes in scope

Hand the packet to whatever you already use. It stays inside the boundaries instead of wandering across the codebase.

Claude CodeCursorAiderCodex

You verify with proof

Run the validation commands Sembl pinned up front. Nothing ships that can't be shown to work.

The Work Order

A contract, not a suggestion.

A prompt hopes the agent does the right thing. A Work Order tells it exactly what it may touch, what it must not, and what "done" actually means.

WO-084 - bugfixRISK MEDIUM

Intent

Send users to the page they originally requested after they sign in, instead of a generic landing route.

Scope

src/auth/redirect.ts
src/auth/session.ts
tests/auth.spec.ts

Acceptance

npm run test:auth passes
no fallback to a generic /dashboard route

Stop if

The failing test can't be located, or the fix would touch routing middleware.

Every Work Order locks eight things before a single line changes. That's what keeps an agent on task.

Intent

The real goal, and the outcome a user would see.

Boundary

Non-goals and areas that stay untouched.

Scope

Editable paths, separated from read-only context.

Context

Files to read first, plus the rules that apply.

Success

Acceptance criteria and what must not regress.

Proof

The exact commands that prove it works.

Safety

When to stop and ask a human.

Executor

An agent-ready prompt that carries it all.

Why it helps

Scope changes the outcome.

Same model, same task. The difference is whether the agent knows where the edges are.

Just a prompt

Unscoped

Files touched14

Changes that landed in scope~40%

With a Work Order

Scoped

Files touched2

Changes that landed in scope100%

Illustrative of a graph-grounded run on a focused bugfix. Real results depend on the task, repo, and model.

Install

pip install sembl. That's the latest.

The latest stable release lives on PyPI, so pip install sembl always gives you the newest version. If you want commits that haven't been released yet, install from GitHub.

Stable from PyPI

pip install sembl

# or run it as an isolated tool
uv tool install sembl

Stable with graph context

pip install "sembl[graph-pipeline]"

# includes Graphify + code-review-graph

Unreleased commits - GitHub

uv pip install "sembl[graph-pipeline] @ git+https://github.com/speedvibecode/sembl.git"

# commits ahead of the PyPI release

Resources

New here? Start with the guide.

Everything you need to go from the install command to your first Work Order - written for people who haven't set this up before.