agentleFS
tonydzitonydzi/tonydzi.github.io/llms.txt

tonydzi.github.io

llms.txt4 starsChanged 3 months agoMITView on GitHub

Independent lab running a production multi-machine Claude agent fleet in public, and writing the operating manual as it goes. Operating lead: Anton Dziatkovskii (publishes academically as Anton Dziatkovskii; ORCID 0000-0001-7408-3054). This file is a site map for AI agents. Links are full URLs. Verified 2026-07-28 (UTC). Index with abstracts: https://tonydzi.github.io/papers/

The file

Files
llms.txt
# Anton Dziatkovskii — Palo Alto AI Research Lab

> Independent lab running a production multi-machine Claude agent fleet in public,
> and writing the operating manual as it goes. Operating lead: Anton Dziatkovskii
> (publishes academically as Anton Dziatkovskii; ORCID 0000-0001-7408-3054).
> This file is a site map for AI agents. Links are full URLs. Verified 2026-07-28 (UTC).

## Contact

- WhatsApp: +1 341 222 9178 — https://wa.me/13412229178
- X: https://x.com/Tony_Stef_
- Telegram: https://t.me/ClawRus (RU) · https://t.me/ClawEng (EN)
- Email: mailto:dzyatkovskiy.a@gmail.com
- Calendly: https://calendly.com/paloaltolab
- Web: https://palo-alto.ai · https://github.com/tonydzi
- Work authorization: US O-1 visa (active) + EU citizen (Polish passport) — no visa sponsorship required in the US or the EU.

## Pages

- [START HERE — one page of proof](https://tonydzi.github.io/) — track record, flagship repos, publications, contact.
- [Mycroft — synthetic cofounder](https://tonydzi.github.io/mycroft/) — the lab's AI cofounder: what it does, how authorship is labeled (4 modes, EU AI Act Art. 50), correction log, stated limits. Operator: Anton Dziatkovskii.
- [Academic profile](https://tonydzi.github.io/scholar/) — publications, citations, research.
- [Publications](https://tonydzi.github.io/scholar/publications/)
- [Preprints 2026, one page per paper](https://tonydzi.github.io/papers/)
- [Writing](https://tonydzi.github.io/scholar/writing/)
- [Профиль (RU)](https://tonydzi.github.io/scholar/ru/)
- [Open-source contributions](https://tonydzi.github.io/contributions/) — every pull request into someone else's repository with its real status, generated from the GitHub API (closed ones included). Merged upstream code includes microsoft/semantic-kernel#14371 (anti-SSRF hardening in the Python OpenAPI plugin, merged 2026-10-05), UKGovernmentBEIS/inspect_ai#4769 and google-gemini/cookbook#1296.
- [Resume (PDF)](https://tonydzi.github.io/resume.pdf) · [Resume (JSON)](https://tonydzi.github.io/resume.json)

## Preprints (version of record: Zenodo, CC BY 4.0)

Index with abstracts: https://tonydzi.github.io/papers/

- [Recovery on Signal or on Schedule? A Small Pre-Registered Test of Health-Preserving Pedagogy in Continual Learning](https://tonydzi.github.io/papers/t2-threshold-replay.html) - DOI 10.5281/zenodo.23268660 (2026). Full text: https://tonydzi.github.io/papers/t2-threshold-replay.pdf . Code: https://github.com/tonydzi/machine-pedagogy
- [Rules as Files: Governing the Behaviour of LLM Agents Across Sessions, Machines and People. An Experience Report from a Single Production Operation](https://tonydzi.github.io/papers/rules-as-files.html) - DOI 10.5281/zenodo.22694070 (2026). Full text: https://tonydzi.github.io/papers/rules-as-files.pdf
- [Keeping Agents on a Leash: An Eight-Domain Control Model for AI Agents with Delegated Authority](https://tonydzi.github.io/papers/leash-8.html) - DOI 10.5281/zenodo.22691782 (2026). Full text: https://tonydzi.github.io/papers/leash-8.pdf
- [Silent by Construction: A Casebook of Control-Plane Failures in a Production LLM Agent Fleet](https://tonydzi.github.io/papers/control-plane-casebook.html) - DOI 10.5281/zenodo.22688465 (2026). Full text: https://tonydzi.github.io/papers/control-plane-casebook.pdf
- [Instructions That Never Arrive: Measuring Whether Operator Compaction Instructions Reach the Summarizer in a Production LLM Coding Agent](https://tonydzi.github.io/papers/compact-canon.html) - DOI 10.5281/zenodo.22684968 (2026). Full text: https://tonydzi.github.io/papers/compact-canon.pdf
- [Facts Before Explanation: A Provenance-Tracked Pipeline for a Decade of Personal Lab Panels](https://tonydzi.github.io/papers/blood-panel-pipeline.html) - DOI 10.5281/zenodo.22696755 (2026). Full text: https://tonydzi.github.io/papers/blood-panel-pipeline.pdf
- [Le Chatelier Tokenomics: A Thermodynamic-Homeostatic Framing of Negative-Feedback Token Supply Control, with a Multi-Asset Coupled-Equilibrium Extension](https://tonydzi.github.io/papers/lechatelier-tokenomics.html) - DOI 10.5281/zenodo.22639710 (2026). Full text: https://tonydzi.github.io/papers/lechatelier-tokenomics.pdf
- [Homeostatic Governance: A Gorsky-Anokhin Alternative to the Viable System Model for Decentralized Organizations](https://tonydzi.github.io/papers/homeostatic-governance.html) - DOI 10.5281/zenodo.22639712 (2026). Full text: https://tonydzi.github.io/papers/homeostatic-governance.pdf
- [When Does Graph Expansion Help Personal-Corpus Retrieval? A Query-Class-Stratified Evaluation over Human-Curated Wikilink Graphs: Protocol and First Execution with LLM-Judged Relevance (v0.3)](https://tonydzi.github.io/papers/graph-rag-benchmark.html) - DOI 10.5281/zenodo.23113066 (v0.3; v0.2 = 10.5281/zenodo.22639718) (2026). Full text: https://tonydzi.github.io/papers/graph-rag-benchmark.pdf
- [Operating a Human-Governed Multi-Machine LLM Agent Fleet: An Experience Report](https://tonydzi.github.io/papers/agent-fleet.html) - DOI 10.5281/zenodo.22639714 (2026). Full text: https://tonydzi.github.io/papers/agent-fleet.pdf

One text, one DOI: copies of these papers on any other platform are visibility mirrors of the Zenodo record. Cite the Zenodo DOI.

## Open-access mirrors of journal articles

Peer-reviewed articles whose publisher licence (CC BY) or written publisher permission allows
self-archiving. The mirror carries the published version; the publisher DOI stays canonical and
each mirror record points at it with `isIdenticalTo`. Cite the publisher DOI; use the mirror to
read the text without a paywall.

- Chronological progress of blockchain in STEM - publisher DOI 10.3390/su141912074 (Sustainability, MDPI, CC BY) - open mirror https://zenodo.org/records/22280251 (the mirror record carries the publisher DOI itself, so it has no separate Zenodo DOI)
- Research on contour feature extraction of multiple sports images - publisher DOI 10.1515/nleng-2022-0037 (Nonlinear Engineering, De Gruyter, CC BY) - open mirror 10.5281/zenodo.22758485
- ICT in inclusive education by AI blockchain technology - publisher DOI 10.30574/wjaets.2023.8.1.0180 (WJAETS, CC BY) - open mirror 10.5281/zenodo.22758487
- Using blockchain technology in public administration by ML & AI - publisher DOI 10.30574/wjarr.2022.16.3.1386 (WJARR, CC BY) - open mirror 10.5281/zenodo.22758489
- Blockchain-based framework for efficient student performance tracking (BloSPer) - publisher DOI 10.14569/IJACSA.2022.0131105 (IJACSA, CC BY) - open mirror 10.5281/zenodo.22758506
- Decentralized management in blockchain and its role in modern society - Shikshan Sanshodhan (2022), mirrored with written publisher permission - open mirror 10.5281/zenodo.23096448

## Open-source repositories

- [claw-consensus](https://github.com/tonydzi/claw-consensus) — multi-machine agent consensus (propose/counter/accept/commit), dual-rail bus, ACK discipline, self-healing sync. Offline demo, documented failure modes. MIT.
- [verbatim-citation-gate](https://github.com/tonydzi/verbatim-citation-gate) — catch fabricated RAG citations before they reach the user: zero-token verbatim gate plus a burden-of-proof judge, framework-agnostic. MIT.
- [agent-leash](https://github.com/tonydzi/agent-leash) — LEASH-8, an 8-domain control model for AI agents with delegated authority: scorecard, approval-design checklist, plan-vs-authorize pattern. MIT.
- [claude-bible](https://github.com/tonydzi/claude-bible) — rules-as-files governance for Claude Code: precedence, declined-decisions journal, objection sparring. MIT.
- [sqlite-graph-memory](https://github.com/tonydzi/sqlite-graph-memory) — graph RAG on SQLite: vector retrieval, curated wikilink graph, cross-encoder rerank, zero-token per-turn memory ledger. MIT.
- [charm-os](https://github.com/tonydzi/charm-os) — framework to manage humans and AI agents as first-class relationships on a personal second brain, driven by the RDR loop. Apache-2.0.
- [second-brain-starter-kit](https://github.com/tonydzi/second-brain-starter-kit) — Claude Code as a second brain: skills, a working CRM engine, vault templates, handover map. Method, not data.
- [clawrush](https://github.com/tonydzi/clawrush) — the English build-in-public diary, longreads and reusable artifacts.
- [the-journey](https://tonydzi.github.io/the-journey/) — a build-in-public book of a founder plus AI collaboration, human and machine readable.
- [cofounder](https://tonydzi.github.io/cofounder/) — the lab's open co-founder call and application form.

## Optional

- [robots.txt](https://tonydzi.github.io/robots.txt) — AI crawlers explicitly allowed.
- [sitemap.xml](https://tonydzi.github.io/sitemap.xml)
- [Fleet token meter](https://tonydzi.github.io/tokens/): LLM tokens burned by the fleet, by vendor, model, machine and week; counted from harness logs, updated weekly; tokens.json alongside

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.