Your decisions deserve a durable form.
You make dozens of strategic decisions inside AI conversations — pricing pivots, hiring calls, market bets. Then the conversation ends, and the reasoning disappears. Three months later, you remember what you decided but not why.
alignmink-dtp is a free, local-first utility for capturing those decisions as structured traces and reading them back in a journal-style interface. The capture skill runs inside Claude. The Decision Journal reader runs on your machine. Nothing leaves localhost.
Traces conform to the Alignmink Decision Trace Protocol (DTP) v0.1 — an open schema you can walk away with at any time.
The simplest path — no install, runs the latest version every time:
npx alignmink-dtp install-skills
npx alignmink-dtp startThe first command activates the capture skill in Claude Code. The second launches the Decision Journal at http://localhost:3000. Set a passcode. Point the reader at any folder containing alignmink-traces/threads/*.json. Sample decisions are pre-seeded.
Prefer a one-time install? See Updating below.
Works with Docker too.
The CLI does three things. If you only learn one command, learn doctor.
| Command | What it does |
|---|---|
alignmink-dtp doctor |
Tells you whether your install is current and prints the literal command to fix it if not. Run this first when anything looks off. |
alignmink-dtp install-skills |
Installs the Claude Code capture skill into ~/.claude/skills/alignmink-dtp/. Re-run with --force to refresh. Your previous skill folder is backed up to <path>.bak-<timestamp> before overwrite, so hand-edits aren't lost. |
alignmink-dtp update |
Updates the Claude Code skill to match the installed CLI. If the CLI itself needs updating, prints the npm command for you. |
There are three independent versions. Most updates touch only one of them.
| Version | Where it lives | Bumped when |
|---|---|---|
| Package | npm release (npm view alignmink-dtp version) |
A new CLI feature ships |
| Skill | engine/SKILL.md frontmatter, copied to ~/.claude/skills/alignmink-dtp/ |
The Claude Code capture skill changes |
| Protocol | schema_version field inside every trace file (DTP) |
The trace JSON shape changes (rare — would break old files) |
If you used npx:
npx alignmink-dtp@latest install-skills --forceIf you installed globally (npm install -g):
npm install -g alignmink-dtp@latest && alignmink-dtp install-skills --forceIf you're not sure which you used:
alignmink-dtp doctorIt detects how you installed and prints the exact command. Run that. Done.
- Decision Journal reader — localhost Node + React app that watches your traces folder and renders decisions as scannable cards with topic / collaborator / time browse lenses.
- Core capture skill — Claude Code skill that detects decision moments in conversation and prompts capture as DTP-conformant JSON.
- Enabler skills (bundled, optional) — War Cabinet (5-advisor pre-commitment panel) and Feedback Panel (post-commitment reflection). Not essential to DTP; useful starter content.
- DTP v0.1 schema — open, Apache-2.0-licensed JSON schema. Read
PROTOCOL.mdfor the spec. - Sample traces — 3-4 realistic decisions loaded on first run so the UI has something to show.
| Surface | Capture quality | Install |
|---|---|---|
| Claude Code | ✅ Premier — silent, zero overhead | npx alignmink-dtp install-skills |
| Cursor / Windsurf | ✅ Direct | npx alignmink-dtp install-skills --target=cursor |
| Claude Cowork | npx alignmink-dtp install-skills --target=cowork |
|
| Claude Chat | (use Cowork or ChatGPT block) | |
| ChatGPT | npx alignmink-dtp install-skills --target=chatgpt |
|
| CLI (non-AI fallback) | 🚧 Coming soon | npx alignmink-dtp capture |
git clone https://github.com/rajconnects/alignmink-dtp.git
cd alignmink-dtp
docker compose up- You have a strategic conversation in Claude.
- The core skill detects a decision moment and asks whether to capture it.
- You confirm; Claude produces DTP-conformant JSON and writes it to your traces folder.
- The Decision Journal renders the new trace the moment it lands.
Nothing phones home. No analytics. No accounts.
Minimal DTP v0.1 trace:
{
"schema_version": "0.1.0",
"trace_id": "2026-04-24-pricing-staged-beta",
"title": "Stage beta pricing at $50/mo",
"captured_at": "2026-04-24T14:32:18Z",
"author": { "name": "Arun Raj", "role": "ceo" },
"decision": {
"statement": "Beta pricing will be $50/month for the first 5 users.",
"reasoning": "Keeps the economics justifying manual support while proving willingness to pay.",
"alternatives": [
{
"option": "$20/mo + $50 setup fee",
"rejected_because": "Setup fee introduces friction in a 5-user beta."
}
]
},
"themes": ["pricing", "beta", "unit-economics"],
"revisit_triggers": ["User count exceeds 5"]
}Every trace captures what was decided, why, what else was considered, and the condition that would reopen the call. Full field reference in PROTOCOL.md.
- Node.js 20+ (or Docker)
- Claude Code / Claude Chat / ChatGPT / any AI surface for capture
- A folder to hold your traces — any folder on disk
Everything runs on localhost. No data leaves your machine. No accounts, no cloud, no telemetry. Your reasoning is your IP.
Apache License 2.0 — see LICENSE and NOTICE.
The Alignmink Decision Trace Protocol (DTP) v0.1 is an open standard. Implementations by others are encouraged and supported.