01.
Alignment
"Every AI tool reads the same decisions. Same context. Same answer — across every chat."
AI that remembers what you decided.
Aligned execution across every chat — Claude, Cowork, Cursor.
The Alignmink Decision Trace Protocol is an open JSON standard. Capture a decision once in any AI chat. Every future conversation reads the same context. Same answer in every tool. Contradictions surface before they ship.
● localhost:3000 — Decision Journal
Easy install with Claude, Cursor, or Cowork.
Two steps, under two minutes. The capture skill goes wherever you reason; the journal runs locally on your machine.
+ Step 1 · Install the capture skill
$npx alignmink-dtp install-skills
Copies the capture skill to ~/.claude/skills/alignmink-dtp/. Silent mid-conversation capture — Claude detects decision moments and writes traces to disk without interrupting your flow.
+ Step 2 · Start the journal
$npx alignmink-dtp start
Opens localhost:3000. Set a passcode on first run. That's it.
Or build from source — github.com/rajconnects/alignmink-dtp. Apache 2.0.
Why this matters
Five reasons your AI tools should remember.
02.
Consistency
"Catch contradictions before they ship. The protocol surfaces when a new call conflicts with a prior one."
03.
Ownership
"Your reasoning is your IP — not your vendor's session log. Open format. On your disk. Yours forever."
04.
Transmission
"The next exec inherits why, not just what. New hires onboard against your captured decisions."
05.
Retrieval
"Past decisions, one query away — by you, or by your AI."
Past decisions become present context.
Once captured, your decision traces feed back into every AI conversation. Claude, Cursor, and Cowork reference the same prior calls when reasoning about new work — and surface contradictions when a new direction conflicts with one you've already locked.
- Recall Every conversation reads the same captures. No re-explaining context across chats.
- Alignment Claude, Cursor, Cowork — same context, consistent answers. One source of truth for every conversation.
- Contradiction A new call conflicts with a prior decision? You see the conflict before you commit.
- Portability Switch tools, your decisions come with you. Open format. Apache 2.0.
claude.ai — Alignmink · Executive CoPilot
Three verbs
Capture. Reference. Align.
One decision. Every AI tool. Same answer.
01 / Capture
In your AI chat
Claude prompts when you decide. The JSON lands on your disk — no copy-paste, no extra step.
02 / Reference
In every future chat
Across any AI tool — Claude, Cursor, Cowork — every conversation reads the same captures.
03 / Align
Across every tool
Execution stays consistent with what you've already decided. Contradictions surface before they ship.
The shift
More decisions are made in chat. Fewer are remembered across them.
Ephemeral by design. Chat windows expire. Reasoning goes with them.
The session ends. The decision is gone.
No tool-to-tool memory. Claude doesn't know what you decided in ChatGPT. Cursor doesn't know what Claude resolved.
Five AI tools. Five blank slates.
One open format. Every DTP-compatible reader works with your traces — today and in five years.
Apache 2.0. JSON. Yours forever.
How it works
Three commands. Under two minutes.
01 / Install
Drop the capture skill into Claude.
One command installs the skill into ~/.claude/skills/. Claude now knows how to detect decision moments mid-conversation.
[ install ]
02 / Capture
Reason out loud. Claude prompts.
When you make a decision, Claude asks: "Should I capture this as a decision trace?" One yes, and the JSON lands under ~/alignmink-traces/.
[ capture ]
03 / Reference
Used by every AI chat.
Every future Claude, Cursor, or Cowork conversation reads from your traces. Run alignmink-dtp start and open localhost:3000 to read them yourself — but the AI is already using them.
[ journal ]
Introducing the Decision Schema.
A memory framework designed to capture the essence of leadership decisions — and an open standard you can read, implement, and extend.
Decision trace · noun
A structured, portable record of a single decision — what you chose, what else was on the table, why, and what would change your mind.
{
"trace_id": "2026-04-30-pricing-staged-beta",
"title": "Stage V1 pricing — $50 beta → $100 launch", ← what you're calling it
"decision": {
"statement": "V1 ships at $50/mo for 5 users; $100 + $20/user at launch.", ← what you chose
"alternatives": [
{ "option": "Single $50/build + $20/mo recurring",
"rejected_because": "Doesn't capture ongoing value once compile is recurring." } ← what else
]
},
"revisit_triggers": [
"If churn at $100 exceeds 30%.",
"If a competitor lands at <$50/mo with comparable depth." ← changes the call
]
}
You never write the JSON. The capture skill prompts you mid-conversation; Claude writes the file. Apache 2.0. JSON Schema 2020-12. Your traces work with any DTP-compatible reader, today or in five years.
Start now
Decisions that outlive the chat.
Aligned execution across every AI tool you use.