OpenTag turns an existing work thread into a governed agent work loop. The CLI configures the local dispatcher, platform listener, runner, executor capability checks, context packet snapshots, action receipts, artifacts, and local audit/status surfaces that keep that loop source-thread-native and local-first.

Install

npm install -g @opentag/cli@latest

Then run:

opentag setup
opentag service status
opentag doctor

opentag setup walks through the local governed-loop configuration:

Choose a language.
Choose a platform: Lark / Feishu, Slack, GitHub, GitLab, Linear, Telegram, or Discord.
Choose a coding agent: Codex, Claude Code, Cursor, OpenCode, Hermes, OpenClaw, or Echo for local testing, including its executor capability boundary. OpenClaw reports cancel=no until its Gateway can prove tool-process termination.
Configure platform credentials.
Bind the selected project.
Choose how OpenTag should run.

The recommended setup option keeps OpenTag running after the terminal closes. It installs and starts a background service on macOS and Linux. If background service mode is unsupported or you choose terminal mode, use opentag start and keep that terminal open.

For GitHub, GitLab, Linear, or Discord webhook deployments that use an already configured relay, opentag setup --relay https://<relay-host> writes relay mode directly and pairs the local checkout with that relay.

opentag status --run <run_id> shows the local context packet, agent work ledger, produced artifacts, queued delivery intents or activation blocks, and safe next actions without turning the source thread into an agent log stream. Provider outcomes are reported only when backed by the delivery journal or signed provider observations.

opentag status --workstream <workstream_id> shows the workstream state and next action first, followed by recipe budgets, accepted outcomes, and bounded exception detail. Healthy workstreams stay quiet. Add --json to receive structured workstream, recipe, metrics, and evaluation objects for automation.

The factory commands operate the recipe, workstream, and restart-safe admission-batch lifecycle through the configured dispatcher. Create and submit commands accept a complete JSON document from a file or from stdin with --input -. Every command supports --json for automation:

opentag factory recipe create --input recipe.json --json
opentag factory recipe get --id recipe_release --version 1 --json
opentag factory workstream create --input workstream.json --json
opentag factory workstream get --id workstream_release --json
opentag factory batch submit --input batch.json --json
opentag factory batch get --id batch_release --json

# The same create/submit input can be piped through stdin.
cat batch.json | opentag factory batch submit --input - --json

The JSON documents use the public FactoryRecipeSnapshotInput, WorkstreamInput, and WorkstreamAdmissionBatchInput contracts. A workstream contains an explicit bounded set of existing WorkThread members; a batch contains ordered admission items for those members. These commands do not define a task graph or move project-management ownership into OpenTag.

External local runtimes can report lifecycle hooks through the public Hook Ingest Contract: opentag ingest-template --format manifest prints the manifest, and opentag ingest records audit-visible progress or terminal state through runner-scoped auth.

Commands

opentag setup
opentag start
opentag service start
opentag service stop
opentag service status
opentag service logs
opentag status
opentag factory recipe get --id <recipe_id> --version <version>
opentag factory workstream get --id <workstream_id>
opentag factory batch get --id <batch_id>
opentag doctor
opentag config path
opentag config show
opentag platforms
opentag executors

Local Config

OpenTag stores local configuration at:

~/.config/opentag/config.json

The config contains local secrets, so the CLI writes it with private file permissions.

Platform Guides

The setup wizard links to the matching guide for each platform:

Lark / Feishu: docs/platforms/lark.en.md
Slack: docs/platforms/slack.en.md
GitHub: docs/platforms/github.en.md
GitLab: docs/platforms/gitlab.en.md
Linear: docs/platforms/linear.en.md
Telegram: docs/platforms/telegram.en.md
Discord: docs/platforms/discord.en.md

Requirements

Node.js 22.14 or newer.
An existing local login or provider configuration for the selected Codex, Claude Code, Cursor, OpenCode, Hermes, or OpenClaw ACP agent.
Platform credentials for the platform you connect.

No Install

The scoped CLI package supports one-off runs without a global install:

npx @opentag/cli doctor
npx @opentag/cli setup
npx @opentag/cli start

For background service mode, install the CLI globally first so the generated service definition points at a stable CLI path instead of an npx temporary location.