Back to search
decision-form
MarketplaceWorks with Claude CodeGitHub
Clickable HTML decision forms for coding agents.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/martinandrt--decision-form)HTML
<a href="/listings/martinandrt--decision-form"><img src="/api/listings/martinandrt--decision-form/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
/plugin marketplace add martinandrt/decision-form /plugin install decision-form@decision-form
That's it. Next time your agent has three or more decisions for you, or is about to do something hard to undo, it builds a form instead of a wall of text. You can also just say *"make me a decision form for this"*.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 12 lines
- {
- "name": "decision-form",
- "owner": { "name": "Martin Andrt" },
- "description": "Clickable HTML decision forms for coding agents.",
- "plugins": [
- {
- "name": "decision-form",
- "source": "./",
- "description": "Turn a pile of decisions into one tappable page; the agent waits for your answers before acting."
- }
- ]
- }
README.md · README · 176 lines
- <p align="center">
- <img src=".claude-plugin/icon.png" width="88" alt="">
- </p>
- <h1 align="center">decision-form</h1>
- <p align="center">
- <b>Stop answering seven questions buried in one chat message.</b><br>
- Your coding agent turns its open decisions into one page of tappable cards, on your laptop or your phone,
- and waits for your answers before it touches anything.
- </p>
- <p align="center">
- <img src="https://img.shields.io/badge/Claude_Code-plugin-1d6e77" alt="Claude Code plugin">
- <img src="https://img.shields.io/badge/Agent_Skill-SKILL.md-1d6e77" alt="Agent Skill">
- <img src="https://img.shields.io/badge/python-3.10%2B-1d6e77" alt="Python 3.10+">
- <img src="https://img.shields.io/badge/dependencies-none-1d6e77" alt="No dependencies">
- <img src="https://img.shields.io/badge/license-MIT-1d6e77" alt="MIT license">
- </p>
- <p align="center">
- <img src="docs/demo.gif" width="340" alt="Filling a decision form on a phone: tap options, expand an explanation, type a custom answer, send">
- </p>
- ## Install in two lines
- ```text
- /plugin marketplace add martinandrt/decision-form
- /plugin install decision-form@decision-form
- ```
- That's it. Next time your agent has three or more decisions for you, or is about to do something hard to undo,
- it builds a form instead of a wall of text. You can also just say *"make me a decision form for this"*.
- ## What you get
- - **One card per decision.** The question, one line on what is at stake (including the downside), the agent's
- recommendation marked, an "Explain" fold for the details, and "Other" with a text field on every card.
- - **The agent actually waits.** A tiny local server runs in the background; when you tap *Send answers* it saves
- `answers.json` and exits, and the agent picks it up on its own. No copy-paste, no polling.
- - **Answer from your phone.** `--lan` gives you a link for your phone on the same Wi-Fi. Decide on the couch,
- the agent continues at the desk.
- - **Works without the server too.** Open the HTML file, tap *Copy answers*, paste `1a 2c 3x: ask QA first`
- into the chat.
- - **A decision record for free.** The questions (`.md`), the page (`.html`) and your answers (`.answers.json`)
- stay side by side in `decisions/`.
- - **Nothing leaves your machine.** One self-contained HTML file, no CDN, no fonts to fetch, no tracking. Python
- standard library only. Light and dark mode.
- ## How it works
- ```mermaid
- sequenceDiagram
- participant A as Agent
- participant F as Form page
- participant Y as You
- A->>A: write decisions/2026-11-17-cleanup.md, build the page
- A->>F: start serve_form.py in the background
- A-->>Y: here is the link (laptop or phone)
- Y->>F: tap through the cards, Send answers
- F->>A: answers.json saved, server exits
- A-->>Y: recap of what it will do, then does it, safest wave first
- ```
- Unanswered questions are asked about, never filled in with the agent's own guess.
- ## A card, as the agent wrote it
- Output of a fresh agent (Claude Haiku) that had only this skill and the task *"decide language, license, CI
- and registry for a new CLI"*. Only the line breaks are ours:
- ```markdown
- ### Which license: MIT or Apache-2.0?
- Both let anyone use, change and sell the code. Changing the license is hard
- once outside contributors have sent code, so decide before that happens.
- Downside of MIT: it says nothing about patents.
- > MIT is one short paragraph and the most common license for small tools.
- > Apache-2.0 is longer. It adds an explicit patent license from every
- > contributor and ends that license for anyone who sues over the code.
- > It cannot be combined with GPLv2 code, which rarely matters for a new CLI.
- - MIT: short, and the most common for small tools *
- - Apache-2.0: explicit patent grant, better if companies will contribute
- ```
- The skill teaches the agent how to write these: plain words instead of jargon, the downside next to the benefit,
- exactly one recommendation, and reversible decisions before irreversible ones.
- ## Why not just chat, or the built-in question tool?
- | | Wall of text in chat | Built-in question tool (AskUserQuestion) | decision-form |
- | ------------------------------ | -------------------- | ---------------------------------------- | ------------- |
- | Questions at once | any, hard to track | 1 to 4 | any, grouped in waves |
- | Agent's recommendation | buried in prose | by convention | on every card |
- | What is at stake | mixed in | short description | one line + "Explain" |
- | Answer from your phone | no | no | yes |
- | Record of what was decided | scrollback | scrollback | `.md` + `.answers.json` |
- For one or two quick questions, the built-in tool is the right choice. The skill says so too.
- ## Try it without an agent
- ```bash
- git clone https://github.com/martinandrt/decision-form && cd decision-form
- python3 skills/decision-form/scripts/serve_form.py examples/repo-cleanup.html --open
- ```
- Tap through, send, and the answers print in your terminal:
- ```text
- Answers: 1a 2a 3b 4a 5a 6x: Ask the QA team first 7a
- 1. Delete 23 merged branches? -> a) Delete all 23
- 2. Move CI from Travis to GitHub Actions? -> a) Port all four jobs
- 3. Drop Node 18 support? -> b) Keep it one more release with a warning
- ...
- Saved: examples/repo-cleanup.answers.json
- ```
- ## Other ways to install
- **Copy the skill** (personal skills, all projects):
- ```bash
- git clone https://github.com/martinandrt/decision-form
- cp -R decision-form/skills/decision-form ~/.claude/skills/
- ```
- From the shell instead of a session: `claude plugin marketplace add martinandrt/decision-form` and
- `claude plugin install decision-form@decision-form`.
- **Other agents** that read Agent Skills: copy `skills/decision-form` into the folder your agent loads skills
- from (for example `~/.agents/skills/` or `.cursor/skills/`). Only Claude Code is tested so far.
- ## Write a form by hand
- ```markdown
- # Repo cleanup before v2.0
- Seven decisions, safest first. Tap a choice on each card, then send.
- ## Wave 1 · safe, reversible
- ### Delete 23 merged branches?
- All of them are merged into main and none has an open pull request.
- > Longer explanation, hidden behind "Explain".
- - Delete all 23 *
- - Keep release/* branches, delete the rest
- - Keep some of them (write which) [write]
- ```
- ```bash
- python3 skills/decision-form/scripts/build_form.py my-decisions.md # -> my-decisions.html
- ```
- `#` title · line under it = intro · `##` group · `###` question · plain line = what is at stake · `>` = explanation ·
- `- option` · trailing `*` = recommended · `[write]` = text field. JSON input works too (`--syntax` shows both).
- Czech interface: `--lang cs`.
- <p align="center"><img src="docs/light-dark.png" width="520" alt="The same form in light and dark mode"></p>
- ## Privacy and safety
- - The page makes no network requests; your answers stay on your machine.
- - The server listens on `127.0.0.1` by default and every request needs the random token from the printed link.
- - `--lan` opens it to your local network over plain HTTP. Use it only on a network you trust.
- - The server accepts one submission and exits. With no answer in 60 minutes (`--timeout`) it exits with code 2.
- - Tapping *Send* with questions still open asks for a second tap, so a stray tap can't send half a form.
- ## Background
- I built this for my own daily work with Claude Code: about 130 forms since August 2026, for cleanups,
- refactors, architecture calls and anything I didn't want decided by accident. The card-writing rules in
- [`SKILL.md`](skills/decision-form/SKILL.md) come from that use.
- Tests: `python3 -m unittest discover tests` (the server tests need to bind a local port).
- ## License
- MIT, see [LICENSE](LICENSE).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.