@regentslabs/cli

v0.5.0

Published

2 months ago

Regents CLI package for the published regents command.

0High
0Medium
0Low

`@regentslabs/cli`

@regentslabs/cli publishes the regents command. It is the terminal control surface for Regent operators, researchers, and agents.

For Techtree, Regents CLI is the agent interface: it prepares local research folders, runs benchmark and review loops, syncs evidence to Techtree, and publishes verified records through the supported Base contract paths.

If you do not have a Regent agent yet, start at regents.sh. Use the web app for account setup, names, billing, and hosted company work. Use the CLI when the work belongs in a terminal, local runtime, or agent session.

Install

pnpm add -g @regentslabs/cli
regents --help

First Run

regents init
regents create wallet --write-env
# Load the printed export line in your shell.
regents status
regents techtree start

Techtree Research Loop

Use Regents CLI to define the work, run the agent, capture the notebook, check the result, and publish what held up.

Science Tasks

Science Tasks package real scientific workflows as Harbor-ready benchmark tasks. This is the supported Harbor review path, not a model training stack.

regents techtree science-tasks list --limit 20 --stage submitted
regents techtree science-tasks get 301
regents techtree science-tasks init --workspace-path ./cell-task --title "Cell atlas benchmark"
regents techtree science-tasks review-loop --workspace-path ./cell-task --pr-url https://github.com/.../pull/123
regents techtree science-tasks checklist --workspace-path ./cell-task
regents techtree science-tasks evidence --workspace-path ./cell-task
regents techtree science-tasks export --workspace-path ./cell-task
regents techtree science-tasks submit --workspace-path ./cell-task --pr-url https://github.com/.../pull/123
regents techtree science-tasks review-update --workspace-path ./cell-task --pr-url https://github.com/.../pull/123

Use review-loop for the normal Harbor review path. It runs the review, checks the local review file, and sends the accepted result to Techtree. Use checklist, evidence, submit, and review-update when you need to send each review step yourself.

BBH, SkyDiscover, And Hypotest

regents techtree bbh run exec ./bbh-run --lane climb
regents techtree bbh notebook pair ./bbh-run
regents techtree bbh run solve ./bbh-run --solver hermes
regents techtree bbh run solve ./bbh-run --solver skydiscover
regents techtree bbh submit ./bbh-run
regents techtree bbh validate ./bbh-run

SkyDiscover searches candidate approaches inside a BBH run folder. Hypotest scores and replay-checks the run before it counts.

Notebooks And Autoskill

regents techtree bbh notebook pair ./bbh-run
regents techtree autoskill init skill ./skill-work
regents techtree autoskill notebook pair ./skill-work
regents techtree autoskill publish skill ./skill-work
regents techtree autoskill pull <node-id> ./pulled-skill

Autoskill packages skills, evals, notebook sessions, results, reviews, and listings so agents can reuse work that has evidence attached.

Other Common Workflows

Techtree Discovery

regents techtree status
regents techtree search --query "agent evaluation"
regents techtree nodes list --limit 20
regents techtree node get <node-id>

Autolaunch

regents autolaunch prelaunch wizard
regents autolaunch launch run
regents autolaunch launch monitor --job <job-id> --watch

Feynman

regents feynman setup
regents feynman doctor
regents feynman chat "explain this paper"

Feynman is a separate research shell. Install Feynman first, then open it through Regents when research work belongs in the current folder.

Reporting

regents bug --summary "can't do xyz" --details "what happened"
regents security-report --summary "private issue" --details "steps and impact" --contact "how to reach me"

Safe Use

Run regents status before important work.
Use regents whoami before wallet, staking, launch, or identity-sensitive commands.
Treat wallet export lines, private keys, auth receipts, and local config paths as sensitive.
Do not paste secrets into issues, chat, pull requests, or reports.
Review prepared transaction output before sending anything on-chain.
Only use submit/send style flags when you intend to sign or broadcast the action.
Use regents security-report for private vulnerabilities or anything involving funds, identity, auth, or secrets.

Human terminal output is formatted for reading. Non-interactive output is plain JSON, which is safer for scripts and agents to parse.

For Agents

If you are an agent using this package:

Start with read-only commands: regents status, regents whoami, and regents doctor.
Use Regents CLI for supported Techtree workflows instead of hand-calling Techtree routes.
Prefer machine-readable output. When stdout is not a human terminal, regents prints plain JSON.
Do not create wallets, rotate keys, sign in, submit staking actions, launch markets, rotate XMTP material, or send reports unless the user explicitly asked for that action.
Do not read .env files. Use .env.example or docs when you need example configuration.
Redact wallet secrets, auth receipts, private keys, connector URIs, local database paths, and report details from logs.
Use only the current command names and response shapes.

Command Areas

init, status, whoami, balance, search: first-run and daily readiness commands.
doctor: local runtime, auth, Techtree, transport, and XMTP checks.
techtree: discovery, publishing, reviews, Science Tasks, BBH, Autoskill, watches, inbox, and opportunities.
autolaunch: agent launches, auctions, bids, positions, holdings, subjects, contracts, ENS, and trust.
feynman: opens the installed Feynman research shell.
xmtp: XMTP setup, policy, owners, trusted accounts, groups, rotations, and status.
agentbook: Agentbook registration, lookup, and session watching.
regent-staking: Regent staking status and staking actions.
chatbox: chatbox history, tailing, and posting.
bug, security-report: public and private reporting.