One self.
Every AI.
The fight moved up a level
First, memory left our heads. Notes, threads, decisions, project history. Now procedures are leaving our hands too. The way you research, write, review, and ship turns into prompts, skill files, and configs.
Three layers are forming. Memory holds what the work knows. Skills hold how the work happens again. An orchestrator runs both. Each one assumes a stable self to attach to. None of them define it.
SelfGraph is that self.
A portable skill still needs a self
“Write in my voice” needs the voice. “Use my source order” needs the order. “Hold my review bar” needs the bar. The skill is the procedure. SelfGraph is the self the procedure plugs into.
Open Skills makes procedures portable. The identity underneath stays assumed, never supplied. SelfGraph fills the gap. One self, read by every skill, in every tool.
One self, four context layers
SelfGraph would hold your professional self in four layers. Each layer references the others, so context stays connected instead of sitting in separate boxes. The first two are solid today. The last two are where the hybrid model adds the most.
Domain Encoding
What you work on and the language of your field. Stack, domain terms, the shape of your problems.
SolidWorkflow Calibration
How you like work done. Format, depth, pace, the standards you hold before a thing counts as done.
SolidBehavioral Relationship
How you and an AI work together. Tone, push-back, when to ask, when to act on its own.
GrowingArtifact Layer
Proof of how you work. Outputs, decisions, and the metadata that makes them legible to any tool.
The OpeningFour ideas worth folding in
The work-package model lists what makes a procedure portable: trigger, tools, permissions, dependencies, memory boundaries, scope, proof, and tested environments. Four parts map straight onto SelfGraph and raise its value.
Work-package metadata for the artifact layer
A skill carries the procedure. SelfGraph would carry the identity half of the same contract — your voice, your source order, your review bar, your boundaries. A skill reads SelfGraph and runs as yours, not as a generic default.
A scope model for identity
Open Skills marks each procedure personal, project, team, public, or private. Apply the same scope to identity facets. Some traits stay personal. Some belong to a project. Some are public. Scope decides what each tool sees.
Proof as a first-class field
Open Skills requires evidence the work happened. SelfGraph records your verification standards. Curl the live endpoint. Confirm real data, not mock fallbacks. Check the table count after a migration. Your bar for done travels with you.
Extraction restraint
Open Skills says no most of the time. Most preferences are noise. SelfGraph promotes only the recurring and non-obvious. This keeps the graph an identity layer, not a memory dump.
Queried, not copied
The goal is an operating layer underneath the tools instead of files trapped inside one of them. For skills, the layer does not exist yet. For identity, SelfGraph is designed to be that layer.
Run the one-question test on any setup: if you had to move this tomorrow, what breaks?
A copied file breaks in many ways. Wrong folder, wrong triggers, missing tools, stale version. A live query breaks in none. The tool connects to the same source and reads the same self. Queried, not copied. Nothing moves, so nothing breaks.
The loop, start to finish
A procedure fires in any tool. SelfGraph supplies the self. The output meets your bar, in Claude, ChatGPT, Gemini, or your own agents.
A skill fires in any tool.
The skill asks SelfGraph for the relevant self.
SelfGraph returns voice, standards, boundaries, and project state.
The skill runs with your context built in.
The output meets your bar, in any tool.
The self underneath the stack
Three of these layers describe the work. One describes the worker. The move is to own identity and connect to skills, not to rebuild them. Compete on the layer no one else holds.
What the work knows. The upstream foundation.
How the work happens again. Portable runbooks.
Runs memory and skills together on one set of rails.
Who the work belongs to and how this person works. The connective self.
What to lock, what to build
Decisions
- →Stay identity-first. Reference skills, do not store or run them.
- →Decide the Orchestrator relationship early. Plug in, complement, or compete.
- →Keep the artifact layer identity-framed. Adopt work-package metadata, hold the framing.
- →Settle the upstream OpenBrain license question. Still open from the audit.
Next Steps
- →Propagate the framing. Repo, MCP manifest, CLAUDE.md, bios.
- →Build the artifact layer on work-package-style identity metadata.
- →Add the extraction restraint bar to the capture pipeline.
- →Publish a LinkedIn article on the hybrid concept.
- →Add an ongoing-initiatives section to the AiGNITE site.