Portable Knowledge Core

This starter kit helps you teach an AI what your Tana system means, save that understanding in a versioned file, and reuse the same system across AI clients.

It does not install a predefined productivity system. Start from either an idea or an existing subsystem: the setup skill designs the smallest owner-approved structure, while the discovery skill interprets structure that already exists. Both produce a client-neutral TANA_SYSTEM.md and small workflow skills.

Start here

1. Fork this repository and connect a trusted AI client to the fork and Tana Remote MCP.
2. Run create-tana-system when starting from “I want a system for X,” or discover-tana-system when useful Tana structure already exists.
3. Review the proposed TANA_SYSTEM.md; correct its meaning and boundaries.
4. Store the approved contract in your own repository.
5. Add small workflow skills that depend on the contract rather than duplicating its architecture.

create-tana-system defaults to a short-round Guided Discovery interview that must be confirmed before architecture is proposed. Ask for Quick Start when you want only a goal, a few examples, and the smallest evolvable first version; its assumptions and unresolved meaning stay explicit.

The included workout example demonstrates the pattern:

• retrieve a plan from Tana;
• capture completed work;
• preview the exact structured write;
• write only after approval; and
• read the result back from Tana.

Repository map

• TANA_SYSTEM.template.md — client-neutral contract template
• skills/create-tana-system/ — approval-gated design and setup from an idea, blank workspace, or partial system
• skills/discover-tana-system/ — architecture discovery and contract generation
• skills/retrieve-workout-plan/ — read-only example workflow
• skills/log-workout/ — approval-gated write example
• examples/build-session/ — sanitized example from a public demo workspace
• examples/from-scratch/ — neutral proposed system package with no Tana writes
• fixtures/ — frozen inputs for rehearsal and testing
• demo/RUNBOOK.md — live event sequence, rehearsal checks, and safe recovery states

Safety

• Start read-only and scope discovery to one workspace or subsystem.
• Keep observations, inferences, owner confirmations, and unresolved questions distinct.
• Never publish workspace IDs, node IDs, credentials, or private content.
• Review generated contracts before sharing them.
• Require explicit approval and readback for writes.

Client entry files

AGENTS.md and CLAUDE.md are lightweight pointers to the shared contract and Skills. They do not guarantee that any client will load repository instructions automatically. Confirm the behavior of the specific GitHub integration, project context, or local checkout you use; otherwise explicitly ask the client to read the files.

See LICENSE. This starter kit is an educational v0, not a backup or migration tool.