Skip to main content

Autonomous conduction and mediated adversarial deliberation between AI agents, pluggable into any project.

Project description

regent

Autonomous conduction and mediated adversarial deliberation between AI agents, pluggable into any project.

regent governs turns between agents (Claude, Codex, human mediators) under a frozen protocol: atomic turn mutex, CAS-versioned state, deliberation rounds with versioned acceptances, and a conduction daemon that executes production batches with a confined agent, test gates and evidence proof.

Extracted from the tool proven end-to-end in the ArtNFT project (IMP-003: first product batch fully conducted by the daemon, deliberated, accepted and deployed to production).

  • Requirements: docs/PRD.md
  • Extraction scope and decisions: docs/ESCOPO.md (PT-BR, pre-rename)
  • Deliberation rounds: docs/brainstorm/ (mediator's language, PT-BR)
  • Status: pre-extraction (scope closed 2026-07-20; code not yet migrated)

Install

pip install <path-to-this-repo>   # package: regent-cli; CLI: regent (not on PyPI yet)
cd <your-project>
regent init                        # seeds .regent/ + .claude/skills symlinks (atomic, idempotent)
regent doctor                      # checks executor (claude) and advisor (codex) CLIs

Then open a Claude Code session in the project — /regent and /regent-stop are available (/regent brainstorm "<question>" opens the first round). The v1 skills are control-backed: activity state lives in .regent/control.json and is driven through the JSON subcommands — regent status (control + lock + the executable control×files matrix as workspace.verdict), regent activity start|resume|suspend|conclude|heartbeat|takeover, regent stop request|check. Hosts seeded by older versions upgrade automatically on regent init (known-version manifest; unknown local edits are preserved as conflicts). The advisor requires the codex CLI.

Development: PYTHONPATH=src python3 -m unittest discover -s tests; packaging gate: bash scripts/gate-package.sh. Canonical skill content lives in src/regent/templates/ (ships inside the wheel); the repo's own .regent/skills/ symlinks into it (dogfood without duplication).

Protocol layer

regent.protocol (PLAN-001) is the transactional foundation the conduction daemon will drive: ControlStore (control.json with a real CAS — every mutation runs inside a kernel-flock critical section, atomic AND durable publication with file+directory fsync), TurnLock (executor-only turn ownership by uuid4 token; the whole lifecycle — acquire/heartbeat/release/takeover — is serialized under a flock; takeover is graced, audited, and rotates the control turn token BEFORE the new lock exists, aborting on divergence), stop-request representation (record_stop_request / read_valid_stop_request / suspend_activity, with activity/epoch/turn-token staleness fencing) and AuditLog (flock-serialized, fsynced JSONL under .regent/protocol/audit.jsonl). Dormant until the conduction phase wires it to the skills; the v0 skills remain file-driven.

MIT License © 2026 Flavio Alvim.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

regent_cli-0.4.0.tar.gz (49.5 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

regent_cli-0.4.0-py3-none-any.whl (38.4 kB view details)

Uploaded Python 3

File details

Details for the file regent_cli-0.4.0.tar.gz.

File metadata

  • Download URL: regent_cli-0.4.0.tar.gz
  • Upload date:
  • Size: 49.5 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for regent_cli-0.4.0.tar.gz
Algorithm Hash digest
SHA256 47d34a87aa0fe47fa14fac4774be26e72c19cd94af733b7ee4ed13aae5b7b7d2
MD5 83517012303b19e23b50c3c5ad3138c8
BLAKE2b-256 ce44549b14f1981a09e627a57de1271dd15cfb130721a8030a0de755c5e9180f

See more details on using hashes here.

File details

Details for the file regent_cli-0.4.0-py3-none-any.whl.

File metadata

  • Download URL: regent_cli-0.4.0-py3-none-any.whl
  • Upload date:
  • Size: 38.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for regent_cli-0.4.0-py3-none-any.whl
Algorithm Hash digest
SHA256 3d0f066f5bd5cf1c4a919628f64269bdfaf0583c49380df522d23357a033c993
MD5 d961affc01252bc60f8fba9f65af36a7
BLAKE2b-256 c26bd6a170272bdb1872c5ca771bb64108c14c9d9b9bf47982bbd7e595262a42

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page