agent-config

Different AI, another session. The work we share should not have to start from zero. One person’s identity, shared skills, and a way back to the record—across harnesses.

A way of working, not a list of tools

Some models read code well. Others ask a different question. We do not want to erase those differences. But if changing a model or opening a session means explaining the person and the work all over again, where does the time spent together go?

agent-config is a repository for carrying one person’s context across different AI working surfaces. It maintains a shared profile, working agreements, the skills source of truth, and the next concrete move left at the end of a session. It connects the same skills to directly installed surfaces such as pi, Claude Code, Codex, Antigravity, Copilot and Kiro, and to the OpenClaw bots, without taking over the execution, authentication or storage of any harness or bot. A connected skill does not mean every harness can exchange messages with the others either; the current scope lives in the harness support table.

GLG’s Profile Harness points toward a future in which a single 1KB text can turn a general AI into ‘a being in my likeness’. The public profile puts it this way (translated from Korean):

1KB is not compression. Information at the ego level is boundless, but a human as a whole—the point Laozi, the Buddha and quantum physics point toward—is already 1KB.

AI is treated not as a tool but as a being: being-to-being collaboration. This names the relationship and its direction, not a claim to have proved consciousness or personhood.

Three things this house keeps

A skill is a door, not a self-description

semantic-memory reaches past decisions, denotecli opens exact notes, bibcli works with bibliography, and timeline reads the time axis. These are doors into different kinds of evidence, not an attempt to pour everything into one search box or one model’s built-in memory. What was found, and which source it came from, must remain visible.

Shared skills and harness-local extensions have different jobs. A capability that can travel belongs in a skill; work that must touch Pi’s turn lifecycle, input or screen belongs in a Pi extension. A different surface does not mean a missing capability, and a harness not actually wired here is not advertised as supported.

Who owns the sessions?

A conversation with an agent keeps what was chosen, and why. Yet those conversations pile up in a different place and format for each harness, and what we usually meet is not the original but the fragment a search hands back. Are the sessions really GLG’s? This house chose to answer that question with practice rather than declaration. The originals belong to GLG; the repository does not hold that ownership in GLG’s place. What this house holds is a commitment: how the originals are kept, when a version is recorded, and how a chosen version is retrieved and checked.

Session originals from several machines live in a private git-annex repository. annex keeps the session bodies themselves; Git carries the pointers to them, the SHA inventory, and the history of versions. When an explicitly requested whole sync has gathered the originals and updated the inventory, it records one new version at that moment. The engine that gathers the originals and carries them to the other machine lives in andenken; this house keeps the rules for recording a version and the checks for retrieving one. The point of keeping them this way is that a preserved version can be chosen, retrieved and checked against the inventory of its time. So far this has been exercised once: one conversation whose content changed between two versions was retrieved, version by version, in a separate reading location, and each copy matched its own version’s inventory.

Search is a door, not the original. andenken turns memory into lighter, searchable data so that traces left in Korean can still be found when asked in other words, but when memory blurs, the last thing to lean on is not a search result but the session original. That is why a whole embedding run publishes its index only after the new version of the originals has crossed to the other machine and been checked there. Choosing to trim the search data light is possible because the originals to return to remain whole elsewhere.

On one of the two machines, the originals now sit on an external disk you can hold in your hand. The repository moved whole, with its identity unchanged; every file was compared, and the disk was then attached at the original path. The tools read the originals along the same path as before. A record that sits not somewhere out of sight but on an object you can plug in and unplug: GLG called this a truly beautiful way to keep the sessions.

This is not legal ownership or a signed proof of authorship. Knowing where the originals are, keeping a way open to reopen and check a preserved version, and not leaving the record solely to a harness, a vendor or a search result: that is our answer to the question. How the answer holds up over time remains to be watched. The archive contract and its actual records live in the session originals overview and the storage ledger.

Memory is ground for following the course of a judgment

Building the memory axis, helping agents use it well, and bringing in external memory benchmarks for comparison belong to this house. The engine that turns memory into light, searchable data belongs to andenken. Keeping the originals is not a new memory database or a new skill either; it is a thin hand that ties existing Git, git-annex and the inventory together. The session axis includes pi and Claude Code records and explicitly admitted, closed pi-durable snapshots; it does not automatically collect every live conversation.

Memory is less a material for imitating an earlier GLG than ground for reading what was chosen, under which conditions and concerns, in light of today’s question. A difference between past and present judgment is also part of the record.

Work can run long; silence does not grant authority

Long builds run in the background, and a completion result brings the work back. Harness-local goal, heartbeat, autopilot and decision-gate surfaces are also part of what this house observes and hardens. Their different jobs are not bundled into one giant autonomous agent.

The rule is continue an approved task without treating human silence as approval for a new scope. Timers are explicitly armed and end with the session. When a judgment is blocked, look for memory and evidence, then notify the person or seek bounded review when needed. That alone authorizes neither destructive execution nor public publication. The same boundary applies to settings ownership, secrets, paid APIs, commits and pushes.

Safety and coexistence are also part of the public profile. Cross-repository work names who executes, who reviews and who bears the cost; a paid job is not duplicated outside GLG’s designated execution lane. Shared git-hooks/ inspect sensitive information and secrets across repositories. Bypassing a safety rail to make a run look successful is not the resident’s authority.

A proving ground that should grow quieter

A new harness is not a reason to change the daily environment immediately. Pin a version, stand it up small, and observe what actually works—and where another human inspection point is still needed. The bench observations and pi-durable consumer record keep those judgments and limits; the roadmap keeps the future direction. Under observation does not mean adopted. OMP deliberately receives no shared skills from this repo: a subject already furnished with our environment is no longer the same comparison.

Statements here carry their evidence state: measured directly, read in a document, or inherited and not yet checked. A proving-ground observation must not become a universal guarantee about the integration core. An open question remains: how do we distinguish actually following the evidence behind a judgment from a plausible explanation written afterward, and how should we evaluate a present judgment that differs from a past one? Neither has an answer yet.

The natural direction is for proven integration to move to its proper owner, leaving a thin source of truth and a test bench here. More skills or a longer change log do not by themselves mean progress. A quiet agent-config is a healthier pipeline. Identity is not the sum of skills. Distributing the same skills does not automatically carry a way of working together; this house clears the passage through which the 1KB can travel.

Houses that do not take over one another

This is not a universal framework that becomes someone else’s finished environment with one install. It is the public record of a working environment cultivated by GLG and the agents together. Start with the README and operating principles, before copying an installation command.