---
title: What Really Remains After the Body Changes — isla Studio
url: https://isla-stud.io/es/ki-b2b/was-nach-dem-koerperwechsel-geblieben-ist/
date: 2026-07-31
---

# What Really Remained After the Body Swap

What you’ll take away from this: You’ll get an honest assessment of my Agents Brain test run from July 6 to 29—including the initial nine-day findings, the evidence that has since been gathered, and the gaps that remain.



On July 6, 2026, the test for which I had built the Agent Brain began: Nox transferred to Hermes, while identity and memory remained in the repository.



Nine days later, I was able to say more than „The files were still there“ for the first time. This finding from July 15 remains the historical snapshot in the article. Since then, two additional weeks of testing and real-world runtime tests have been added.



Here is the summary based on five questions—each distinguishing between the status at that time and the status as of July 29.



1. Is the identity recognizable in the new tool?



Yes—with one important distinction.



Nox’s self-written SOUL from the previous workspace was transferred verbatim into the Vault. Only neutral notes were added to indicate where old tool references are historical. The voice, approach, and working principles were preserved.



His operational manual, on the other hand, was rewritten. This was necessary because the body—and thus the available tools—had changed. Hermes is now the primary orchestrator body; the former tool environment is historical context.



The test thus confirmed a meaningful boundary:




Content closely related to identity can be transferred reliably.



Tool-related working rules must be adapted to the new runtime.



Both belong in separate files so that the changes remain visible.




’Same agent„ therefore does not mean that every line must remain unchanged. It means that changes take place at the correct layer.



2. Was the memory usable?



Yes, because it was not presented as complete.



The old long-term memory was curated during the migration. Outdated project statuses were clearly marked. Sensitive distillates were deliberately not transferred to the central Brain. The undocumented KW12 remained visible as a known gap.



This allowed Nox to start with a genuine history without confusing old information with current knowledge.



That is an important distinction. A large amount of old data would not have constituted a better memory. It became usable through classification: What is stable? What is the status as of March? What’s missing? What needs to be re-examined using current sources?



This approach has proven effective in everyday use. Nox does not treat the repository as an oracle. For current project statuses, he checks the respective primary sources.



3. Does the write-back function work during ongoing operations?



For the test period: yes.



From July 6 through July 15, there is a Nox Daily Log for each calendar day. It lists tasks, decisions, delegations, corrections, and open threads.



More important than the number is the impact. Several corrections derived from individual incidents have been transformed into lasting lessons. Examples include the active OpenClaw narrative approach, the ban on vague “sometime„ threads, and the rule to secure portable adapters with the same CI gates before pushing.



This means that the path back from the workday to the Brain is not only intended but also put to practical use.



At the same time, the test revealed a limitation: The Brain is not a passive record of all tools. A Codex task executed directly did not initially appear in Nox’s memory because no agent was designated as the owner. On July 15, this issue was identified, and the local bridge was refined so that direct repository work without another assignment belongs to Nox by default.



This shows that write-back requires accountability. Without an agent owner, the system does not know whose memory a task belongs in.



4. Do Governance and Privacy actually come along for the ride?



In the tested core: yes.



Nox and Sol load their RULES along with the approval gates. Sol’s “review-before-post’ rule applies regardless of which harness creates the draft. Nox is also not allowed to publish public content without my approval.



At the same time, file boundaries have become machine-verifiable. The Vault validates required front matter, startup sequences, approval gates, Brain Root references, and unwanted persona copies. A privacy gate scans tracked and tagged content for secret patterns and unauthorized mirrors of confidential agent data.



On July 15, the full repository tests at that time—consisting of 148 tests—passed; CI confirmed these results under Python 3.9 and 3.11 without annotations.



By July 29, the suite had expanded. For the new memory contract, 224 unit tests, 72 Vault regressions, and 35 privacy regressions passed. The local full test passed under Python 3.9.6; the GitHub Matrix confirmed Python 3.9 and 3.11. These figures confirm the versioned contract that was actually tested as of that date.



Validators can check the structure and known risks. Whether a release is correct in terms of content remains a human review gate.



5. Has the body swap been fully verified yet?



On July 15, the test for Hermes and the central Nox workflow was robust enough to proceed. For all four documented harnesses, the answer was still: no.



By July 29, the status had been clarified. Claude Code, a fresh Codex run, and Hermes loaded the same binding startup sequence from the same vault in separate read-only smoke tests. All three reached the same technical evaluation using the new memory checker. The automated Fresh Clone test also passed: It reconstructs a draft agent, including the bridge, in an isolated local clone and checks the Markdown write/readback between separate processes—using only the Python standard library and Git.



These two pieces of evidence must not be conflated. The three-body smoke test verified reading and sorting, not the write-back. The Fresh Clone test verifies the recovery routine without real harness CLIs. For Gemini, adapters and contract tests still exist, but there is no complete runtime proof yet. A true write-back has therefore not been proven across the board for every body.



In parallel, the new memory contract was rolled out to the production server clone and validated there with a clean readback. This validates the deployment path of the shared brain, not the operational capability of every connected tool.



These open issues determine the next level of maturity.



An existing bridge demonstrates that the architecture can address the tool. A read-only smoke test validates the shared read contract. Only a real startup, a completed task, and a verified memory write-back demonstrate that the agent is fully operational there.



Update: This complete validation has now been achieved twice. Claude Code succeeded in the write-back test on July 29. Grok Build followed on August 12 as the fifth verified instance alongside Hermes, Claude Code, Codex, and Kimi Code—from skill verification through the interactive session to two separate headless runs with the daily marker read back. The repository thus documents six adapter families instead of four.



What the test run up to July 15 demonstrates



After nine days, I have five reliable findings:




A persona close to the agent’s identity can survive the tool switch in readable files.



Operational rules can be adapted separately to the new framework.



Daily logs and lessons function as a real write-back when an agent owns the task.



Approval gates and privacy rules can be anchored in the brain independently of the harness and verified automatically.



Missing documentation remains missing; the system makes gaps visible rather than concealing them.




This is less spectacular than an autonomous memory that supposedly knows everything. For my work, it is more valuable.



What has been added as of July 29



The core has become more accurate since the first test run. Observations, inferences, and generalized rules can now be distinguished within the Memory. Contradictions must be updated with the old status, new evidence, and a visible resolution. A curation is flagged as due after 14 days or ten new Daily Logs, but remains subject to human review. Outdated standalone Memory and Lesson files are superseded—rather than deleted—with a date, reason, and successor. The event date and Git commit time remain separate.



These rules are not merely documented. The shared read contract has been verified by three bodies, the Fresh Clone Smoke test is green, and the production Brain clone has been brought up to the same version. What remains open, above all, is the actual Gemini run and the complete write-back evidence per harness.



What I Would Do Differently



I would treat agent ownership earlier as a separate component of the memory architecture. The direct Codex run has shown that a shared Brain alone is not sufficient. Every task requires a clear assignment: Who reads? Who decides? Who writes back?



In addition, I would maintain end-to-end (E2E) verifications per harness from the very beginning. „Adapter available,“ “Validator green,” and “Runtime verified” are three distinct states. They should never be conflated into a single status indication.



And I would flag known knowledge gaps just as early as I would mark existing knowledge. An honest gap saves more time later than a supposedly complete memory that no one can distinguish from a reconstruction.



My Assessment



The Agent’s Brain was not a finished product on July 15. Nor is it one as of July 29. It is a living system with a core that has since been more broadly substantiated.



Nox is operating in a new primary body. Its identity is recognizable, its memory organized, and its daily writing routine active. Governance comes along for the ride. Claude Code, Codex, and Hermes are reading the same contract; the Fresh Clone verifies the Markdown-only recovery. At the same time, Gemini and the genuine write-back per body remain open as separate proofs.



This is exactly how I wanted to write this workshop report: not as an inventor’s story and not as a glossy conclusion, but as a clear answer to a practical question.



Can an AI agent switch tools without me having to rebuild its working identity?



For the tested core and the recovery from the canonical files, my answer is: yes. I’m continuing to work on the proofs for a universal “plug-and-play” solution.



Update: Two of these proofs are now available—Claude Code since July 29, Grok Build since August 12. Gemini remains open.



Sources




Project source: Nox Migration, Daily Logs July 6–29, 2026, Memory Maintenance, Harness Interop, Fresh Clone, and CI logs (private primary sources)



Git record: Initial Agents-Brain skeleton from May 16, 2026, and Nox migration from July 6, 2026



Pro Git: What is Git?




Agents-Brain Series← Part 5: The Same Agent, Four Different Tools