alignmink-dtp

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.

60-second quickstart

The simplest path — no install, runs the latest version every time:

npx alignmink-dtp install-skills
npx alignmink-dtp start

The 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.

Three things to remember

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. |

Updating

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 --force

If you installed globally (npm install -g):

npm install -g alignmink-dtp@latest && alignmink-dtp install-skills --force

If you're not sure which you used:

alignmink-dtp doctor

It detects how you installed and prints the exact command. Run that. Done.

What you get

• 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.md for the spec.
• Sample traces — 3-4 realistic decisions loaded on first run so the UI has something to show.

Supported surfaces

| 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 | ⚠️ Download JSON + upload to reader | npx alignmink-dtp install-skills --target=cowork | | Claude Chat | ⚠️ Download JSON + upload | (use Cowork or ChatGPT block) | | ChatGPT | ⚠️ Download JSON + upload | npx alignmink-dtp install-skills --target=chatgpt | | CLI (non-AI fallback) | 🚧 Coming soon | npx alignmink-dtp capture |

Docker

git clone https://github.com/rajconnects/alignmink-dtp.git
cd alignmink-dtp
docker compose up

How capture works

1. You have a strategic conversation in Claude.
2. The core skill detects a decision moment and asks whether to capture it.
3. You confirm; Claude produces DTP-conformant JSON and writes it to your traces folder.
4. The Decision Journal renders the new trace the moment it lands.

Nothing phones home. No analytics. No accounts.

What a trace looks like

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.

Requirements

• 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

Privacy

Everything runs on localhost. No data leaves your machine. No accounts, no cloud, no telemetry. Your reasoning is your IP.

License

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.