Decision Packet Writer
Turns a decision you have been circling into one page someone can act on. Recommendation first, facts and assumptions in separate columns, and the one thing that would change the call.
Free. View purchase detailsBuilt for: A founder or marketing lead with a decision that has been open for more than a week.
Your first deliverable
a decision packet with facts, assumptions and a recommendation your team can review.
A decision someone can actually answer
Worked example in the pack · illustrative decision
The included packet makes one recommendation and shows the reader exactly what it depends on.
- Recommendation
- One proposed action appears before the supporting analysis.
- Evidence
- Known facts and untested assumptions sit in separate columns.
- Change the call
- A specific condition and timeframe state what would reverse the recommendation.
Based on examples/worked-example.md. The example's numbers are illustrative, not measured customer results.
Included in the workflow
- The agent
- A full worked packet
- Setup interview
- Scorecard
- No-terminal project file
Check the fit before buying
A founder or marketing lead with a decision that has been open for more than a week.
- An AI assistant that can read the supplied files
- A real decision and the facts you have so far
Read the package documentation
What follows is the README from inside the zip, unedited. The page and the package say the same thing on purpose.
Free. Version 1.1.0.
Turns a decision you have been circling into one page someone can act on.
Why this is free
It is the smallest complete thing on the shelf and the easiest to judge. Run it once on a decision you have genuinely been stuck on. If the page it produces is not better than what you would have written, none of the paid analysts will be either, and you should not buy them.
That is the entire pitch. There is no limited version here; this is the whole agent.
What is inside
| File | What it is |
|---|---|
agents/decision-packet-writer.md | The agent: procedure, output contract, measurement, guardrails |
examples/worked-example.md | A full packet written to the contract, with notes on what to notice |
SETUP.md | Paste-in interview that picks your first decision and runs it |
scorecard.md | Whether it is working |
projects/ | The no-terminal version for Claude Projects or a custom GPT |
How to use it
- Pick a decision you have been going back and forth on for more than a week. Those are the best candidates, because the delay usually means it was never framed rather than never analysed.
- Read
examples/worked-example.md. - Paste
agents/decision-packet-writer.mdinto Claude or ChatGPT and give it your decision. - Check the assumptions column first. That is where a packet is usually wrong, and it is the part you can correct from your own knowledge in a minute.
The rules it follows
- Recommendation first, one option, never "it depends".
- Facts and assumptions in separate columns, always.
- "Do nothing" is always an option, and it is costed.
- Every option gets its strongest case, including the ones not recommended.
- No invented numbers. A gap is marked, not filled.
- One line on what would change the recommendation.
What it does not do
It does not make the decision, and it does not gather the facts. Given thin inputs it will produce a packet that is mostly assumptions, which is a correct and useful result: it shows you exactly what you need to go and find out.
Structured template and offline writer
Start with START-HERE.md for a ten-minute entry exercise. templates/decision-packet.md is a blank copyable page. templates/packet.json is the structured blank; examples/structured-input.json is an openly fictional runnable case. The original agent, guided setup, project form, scorecard and worked example are all retained.
python3 tools/write_packet.py --input examples/structured-input.json --output my-packet
python3 -m unittest discover -s tests -vThe writer separates checked facts from assumptions and unverified claims, preserves unknown costs and requires the do-nothing option. It generates report.md and report.json locally. No keys or third-party libraries are required. It validates structure and evidence labels, not factual truth. See INPUT-CONTRACT.md. To uninstall, save your own work and remove the downloaded folder; nothing is installed elsewhere.
Before you start
The files
The agent. A full worked packet. Setup interview. Scorecard. No-terminal project file.
Setup requirements
An AI assistant that can read the supplied files A real decision and the facts you have so far
The first task
A one-page packet for the decision you have been stuck on, with the assumptions column you can correct in a minute.
Who runs the work
You install and run the files, inspect their output and decide what to use. A file purchase does not include consulting, setup calls or working sessions.
Payment and updates
These files are free. Marketing emails are a separate, optional choice.
Also in Free starters
Put the workflow to work.
Read the requirements, bring your inputs, and review the result before you use it.