Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Axionary

A system that turns incoming documents into a structured knowledge base of actionable cards, organized by company function.


Why

Every day brings a stream of information: articles, notes, reports, emails, meeting transcripts, market news. Most of it ends up buried in personal folders or lost in chat, never turning into anything concrete.

Axionary solves a simple problem: take any incoming text and extract specific actions from it — with context, rationale, and the risks of ignoring them — filed under the functions of the company they belong to.

The result is a graph of linked cards in state/ that you can navigate: what needs to be done, why, and what connects to what.


How it works

Directory structure

axionary/
├── inputs/       ← drop incoming documents here
├── processed/    ← processed files are moved here
└── state/        ← actionable cards live here
    ├── strategy/
    ├── product/
    ├── engineering/
    ├── competitive-intelligence/
    └── ... (created on demand)

Processing cycle

  1. Drop a file into inputs/ (markdown, txt, csv, pdf, docx — any text document).
  2. Ask Claude to process it: "process inbox" / "run the next file" / "triage inputs".
  3. Claude runs the inbox-actionizer skill:
    • picks one random file from inputs/ and moves it to processed/
    • reads the content
    • identifies what concrete actions follow from it
    • assigns each action to the most precise vector (function) from a list of ~200
    • searches for associations with existing cards in state/
    • creates cards in the corresponding subdirectories
    • adds bidirectional wiki-links between related cards

Card format

Every file in state/ is markdown with frontmatter and five sections:

---
title: ...
vector: strategy
source: source-file.md
created: 2026-05-17
status: open
related:
  - "[[partnerships/0001-evaluate-channel]]"
---

## Background
What in the document prompted this action.

## Hypothesis
What we believe: if we do X, then Y will happen.

## What to do
Concrete steps.

## Expected positive outcomes
What will improve and for whom.

## Risks of inaction
What will happen if this is ignored.

## Links
- [[another-vector/file]] — why it's related

File names follow the pattern <sequential-number>-<kebab-slug>.md, e.g. 0003-pricing-audit.md. Numbers are monotonically increasing within each subdirectory.

Cross-linking

Cards are linked in both directions: when a new card A references an existing card B, the skill writes the backlink into B as well. This turns state/ from a flat archive into a navigable graph — like an Obsidian Vault.


Usage

Add a document for processing

Copy any text file into inputs/:

cp ~/Downloads/market-brief.pdf inputs/
cp ~/Desktop/meeting-notes.md inputs/

Run processing

In Claude Code (from the project directory), say:

process inbox

or

run the next file

One request = one file. To process everything in inputs/ in a loop, run claude -p in an external loop.

View results

Cards live in state/<vector>/. They are plain markdown and open in any editor. Obsidian works especially well — wiki-links in [[vector/file]] format are native there, and you can view the graph.

Update card status

Open a file and change status: open to status: done (or in-progress, archived — whatever works for you).


Vectors (functions)

The skill recognizes actionable items across ~200 functional areas: from product and engineering to treasury, procurement, trust-and-safety, innovation-labs, and corporate-social-responsibility. The full list is in .claude/skills/inbox-actionizer/references/vectors.md.

Each card belongs to one most precise vector. Adjacency is expressed through links, not duplication.


Technical details

The inbox-actionizer skill lives in .claude/skills/inbox-actionizer/ and contains:

  • SKILL.md — instructions for Claude
  • scripts/triage.py — helper script for atomic file moves and monotonic numbering
  • references/vectors.md — the list of allowed vectors

Requires Python 3 and an AI agent (Claude Code, Cursor, ...).

Connecting an MCP search tool is strongly recommended — for example, Obsidian Hybrid Search.

About

Turns incoming documents into a structured knowledge base of actionable cards

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages