Evidence beside every claim.
Boxes and notes link to the exact words, numbers or lines of code behind them, with the source and date. Customer and patient quotes are marked as their own words.
Decisioncraft
Point it at anything. See how it works, what is missing, and what every role thinks.
Point Decisioncraft at anything (a codebase, meeting notes, a process or a personal choice) and you get a map of as-is against to-be, the gaps written as user stories, and a sticky note from every role. It's rigorous decision analysis in plain language, from a quick side-by-side to a full team review.
Install
The portable archive includes the method, Python library and references. CLI and MCP installation are separate options.
Open a complete example without a model.
Read a fictional decision map, validate its structure and render it to one local HTML file. Explore its claims, options and role notes before asking the tool to map your own work.
A fictional first run
git clone https://github.com/michaeljabbour/amplifier-smart-tool-decisioncraft.git cd amplifier-smart-tool-decisioncraft decisioncraft validate examples/engineering/model.json decisioncraft render examples/engineering/model.json --out first-canvas.html
Install the native tool first, then run these commands in a terminal. Open first-canvas.html in your browser. This is an existing fictional example; no key or provider call is needed. Validation checks structure, not the truth of its claims.
How it works
Point it. Hear everyone. Decide.
Start from what you already have: a repo, notes, interviews, figures or just a question. Decisioncraft turns it into one map that a mixed group can read without training.
Point it at anything.
decisioncraft map ./your-repo reads the code, notes or page, cites every claim by file and line, and draws today's way and the planned way. No model to route? map --starter gives your agent a digest and a model to fill in.
Hear every point of view.
Designers, architects, analysts, engineers, product, security, the people you serve, an AI agent teammate and finance each leave a sticky note beside the boxes. Every note ends in a question. Add your own rough note and ask the experts.
Decide at the right depth.
A quick table in the chat for a small choice, a short interview and a map for a costly one, a full review for a team. Merge reviews to see where people agree and split, then record the decision with an owner and a date.
Made for real work
Keep the reasons with the decision.
What changes, at a glance.
One map marks every box New, Changed or Goes away, with a walk through each change and before and after side by side. Side by side shows today and the plan in two panes.
Scores and costs you can trust.
Must-haves first, weighted scores with what would change the winner computed, and real cost over time with break-even points and what-ifs.
Point it at a repo
A small app, mapped in one go.
decisioncraft map --starter read a made-up bike hire repo; a coding agent filled in the map from it, citing each file and line. Today's way, the planned way, six gaps with user stories, and a sticky note from each of eight roles.
For a personal choice
Keep, lease, buy, or go car-free?
Five options for a family whose lease is ending. Must-haves first, a scoring table with live weights and what would change the winner, and five years of real cost with break-even points and what-ifs. Real published figures, a fictional family.
Ask the experts
Stick a rough note on it. Hear every role.
A shop manager worries that collection only would lose two villages. The customer voice, finance and the owner each reply under the note with a short view and a question, which joins the list of things to decide.
For a business decision
A bakery weighs a monthly box.
Three shops, customer chats, 64 survey slips, a waste log and the till supplier's notes. Six journeys follow a subscriber's month and the work behind the counter, then trace the idea from what customers said to a pilot plan.
For a technical decision
A web team moves its file uploads.
An incident review, a month of upload numbers, support tickets and a new provider's terms. Six journeys show uploads, downloads, deletes, a file move and failures across the system, with an options tree beside them.
For an engineering decision
A town decides on a footbridge.
An inspection, crossing counts, an access audit, quotes for a way across and a shops survey. The canvas traces the decision, compares three options from committee to reopening, and follows a resident through a closure.
For a decision in care
A ward changes how it sends people home.
A worked example about how a ward organises its work, not medical advice. Seven journeys follow a patient and carer home and the ward's evening list, medicines, discharge day and follow-up calls. Everything is invented.
Get started
Bring it to
your agent.
Start with the outcome you want. Your agent can install the tool and use its help to carry out the work.
A starting prompt
Show me how this repo works and what's missing. Use Decisioncraft: read decisioncraft --help, run decisioncraft map . --starter --dir repo-map, fill in the model from the digest using decisioncraft guide, validate it, then render the canvas and open it.
Help your agent can use.
Amplifier Smart Tools provide help in skill format: instructions your agent can read to understand capabilities, inputs, and how to use the tool. Ask your agent to read --help before getting started.
Terminal
uv tool install "amplifier-smart-tool-decisioncraft[smart,mcp] @ git+https://github.com/michaeljabbour/amplifier-smart-tool-decisioncraft" decisioncraft --help
Python 3.11 or later. Canvases are single HTML files that open in any current browser, offline.
map and draft need a model: your host's own (by command or MCP sampling) or Anthropic or OpenAI access. Without one, map --starter and decisioncraft guide let your agent write the map itself. Triage, the interview, quick tables with scores given, drawing, comparing and merging reviews need no model.
Part of the same family
Keep making.
Each tool stands on its own. Explore the others when your work takes you there.