Skip to content

sin2664-collab/alignor

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 

Repository files navigation

alignor.md

A single markdown file that turns any LLM into your alignment guardian.

Paste it into Claude Projects, ChatGPT Custom Instructions, or Cursor rules.
No install. No API key. No setup.


The problem

AI makes you 100x faster.

But if your direction is off by 1 degree, you end up 100x further from where you wanted to be.

You've been there: shipped a ton of stuff, checked every task off the list — then looked up and realized none of it moved the needle on what actually mattered.

This is the Productivity Paradox of the AI age.
Speed without alignment is just accelerated drift.


What alignor does

You define your mission (why you exist) and visions (concrete future states you're aiming for) once.
Every task your LLM generates gets tagged, scored, and checked against your visions.
Drifted items get flagged.

[TASK] Re-engagement newsletter to inactive users
[WHY]  Recovers lapsed users — directly moves V1 forward
[VID]  V1
[OUT]  +15% retention among inactive segment within 30 days
[PRI]  HIGH

[TASK] Reorganize internal Notion workspace
[WHY]  Operational tidiness — no connection to any vision
[VID]  NONE
[OUT]  Cleaner workspace
[PRI]  LOW

─────────────────────────────────────
ALIGNMENT CHECK
─────────────────────────────────────
V1 (grow active user base):  [████████████████░░░░]  82/100  ✓
V2 (launch in new market):   [░░░░░░░░░░░░░░░░░░░░]   0/100  — no tasks this week

MISSION COHERENCE:           [████████████████████] 100/100

DRIFTED:
  ✗ Reorganize Notion — operational, moves no vision forward

VERDICT: PARTIAL
V1 is well-covered. Nothing moved V2 this week — intentional, or crowded out?
─────────────────────────────────────

Mission vs Vision — why both matter

Most tools conflate these. alignor keeps them separate on purpose.

Mission Vision
What it is Why you exist Where you're going
Changes? Never Yes — as strategy evolves
Abstraction High Concrete and observable
Role in alignor Background context The actual scoring target

A task that "feels aligned with the mission" but serves no vision is still DRIFTED.
Visions are the scoring targets. Mission validates that your visions are pointing the right way.


Setup (5 minutes)

1. Download alignor.md

2. Fill in Section 1 — Mission

Use the fill-in-the-blank structure:

We exist to [core purpose].
We believe [fundamental belief others might disagree with].
We will never [what you refuse to do, even for money].

3. Fill in Section 2 — Visions

Use the fill-in-the-blank structure:

V1: By [timeframe], [who] will [do or experience what].
V2: By [timeframe], [who] will [do or experience what].

4. Paste the whole file into your LLM

Done. Every conversation from now on is alignment-aware.


Works with any LLM

Tool Where to paste
Claude Projects System Prompt
ChatGPT Custom Instructions
Cursor .cursor/rules
Windsurf .windsurfrules
Gemini Gems Custom Gem instructions
n8n / MCP First system context node
Any API system parameter

Why a markdown file?

No SaaS to sign up for. No data leaving your machine beyond what you already send your LLM.
No workflow changes. Just a file you own, control, and can edit in 30 seconds.

The file is the product.


FAQ

Isn't this just prompting an LLM?
Yes. The value isn't the technology — it's the structure. Most people have a mission in their head. Few have one legible to their tools. The fill-in-the-blank format forces enough specificity for scoring to actually mean something.

Why score against visions, not the mission?
Mission is too abstract to score against reliably. "Does this task serve our purpose?" — an LLM will say yes to almost anything. "Does this task move V2 forward?" — that's concrete enough to score honestly.

Can't anyone copy this?
Yes. The templates, community configs, and habit of using it are what compound over time.

Does this work with GPT-4 / Gemini / Mistral?
Yes. Tested with Claude, ChatGPT, and Gemini. Any instruction-following model works.


Roadmap

  • v0.1 — single markdown file, works with any LLM
  • v0.2 — CLI: cat output.txt | alignor check

Contributing

Most useful contributions right now:

  1. Your config — what mission/vision format worked for your use case
  2. Template PRs — your industry's version of alignor.md
  3. Failure cases — when did scoring feel wrong? Open an issue.

Speed without direction is just accelerated failure.

MIT License · Made by @yourhandle

About

A markdown file that scores your AI outputs against your mission

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors