general-augment-cli 0.3.0

CLI for General Augment, the agent backend for your app.

General Augment CLI

General Augment is the agent backend for your app. This is the standalone developer CLI for creating, validating, deploying, and verifying General Augment projects, and for wiring up auth, providers, tools, and skills.

uvx --from general-augment-cli genaug --version
uvx --from general-augment-cli genaug launch --inspect --json

uvx is the preferred checkout-free invocation for coding agents. pipx install general-augment-cli remains supported when a persistent installation is preferred. Standard virtual environments may continue to use pip install general-augment-cli.

For source-checkout development, use the repo-local command prefix:

uv run --project packages/cli genaug --version
uv run --project packages/cli genaug --help

The version reported by genaug --version is single-sourced from the installed package metadata, so it always matches the published distribution.

Command surface

The published CLI is intentionally minimal — the setup, integration, and verification commands an app developer needs, and nothing else:

Command group	Commands
genaug auth	login, logout, whoami
genaug keys	list, create, update, revoke
genaug tools	list, toggle, catalog, discovery, explain-turn, add-mcp
genaug skills	design, list, view, apply, delete
genaug providers	setup, smoke, readiness
genaug connectors	setup
genaug agent	list, show, status; delegation list
genaug migrate	openai-responses
genaug dashboard	open
genaug launch-skill	install, status, remove
top-level	launch, integrate, init, setup, doctor, status, smoke, validate, verify

Quick start

For the supported Next.js App Router and Clerk beta lane, start with the read-only machine-readable inspector:

uvx --from general-augment-cli==0.3.0 genaug launch-skill install --agent all --scope project --version 1.2.0 --json
uvx --from general-augment-cli==0.3.0 genaug launch --inspect --json
genaug launch --questions --json
genaug launch --plan --answers-file .genaug/launch-answers.json --json
genaug launch --review --json

--questions is the read-only handoff for humans and coding agents. It reports the authenticated account, available Workspaces and Projects, detected application identity, and unresolved Agent-topology or release decisions using stable question IDs. The answers file is secret-free and can select or create a Workspace and Project, declare multiple Agents, assign Project tools, Skills, and memory, and review delegation edges. Skill assignments require skills_directory to identify the repository-relative Project Skill library; provisioning pins those reviewed files by content hash. The CLI and dashboard use the same hierarchy: Account -> Workspace -> Project -> Agent.

The review command returns the exact dashboard integration-review URL. After approval, provision and explicitly authorize the ignored local environment handoff:

genaug launch --provision --approve-session <session-id> \
  --configure-application-env --json

The CLI applies project configuration with installer authority and stores the separate responses:create runtime key only in its owner-readable config and the verified Git-ignored .env.local. Output contains variable names, paths, and masked key metadata, never values. Re-running reuses the one active key; use --rotate-runtime-key for an explicit revoke-and-replace operation. Rollback can remove only the managed env block:

genaug launch --provision --approve-session <session-id> \
  --remove-application-env --json

Then apply the official generalaugment-launch coding-agent skill and finish with:

genaug launch --verify --json
genaug launch --finalize --configure-application-env --json
genaug launch --open-dashboard --no-browser --json

Release operators can bind the temporary preview-key provisioning proof to the durable runtime key created by finalization:

genaug certification create \
  --workspace . \
  --verification .genaug/launch-verification.json \
  --provision-first .genaug/evidence/provision-first.json \
  --provision-second .genaug/evidence/provision-second.json \
  --finalization-first .genaug/evidence/finalization-first.json \
  --finalization-second .genaug/evidence/finalization-second.json \
  --browser .genaug/application-verification.json \
  --deployment .genaug/hosted-deployment.json \
  --json

Both finalization receipts are required together. The first records creation or reuse of one durable Test key; the second must reuse that exact key. Certification fails when the launch, Project, immutable release, runtime mode, owner-only environment handoff, or deployment identifiers differ. Receipts contain IDs and hashes only, never raw keys.

After a human-approved baseline, a coding agent may ask the server to apply a bounded safe Test change with genaug launch --activate --auto-approve-safe --json. The flag never grants authority: server-owned Project policy either approves the exact fingerprint or returns the dashboard review URL. Inspect current state with genaug agent list|show|status; Agent mutations are made in genaug-agent.yaml and applied through genaug launch so imperative installer commands cannot bypass review.

Set GENAUG_DASHBOARD_URL when an isolated control plane uses a non-production dashboard. Launch review and verification links will target that dashboard instead of https://app.generalaugment.com.

The skill installer copies the versioned official skill into the current project's Codex and Claude Code skill directories. It verifies every file against the SHA-256 manifest bundled in the CLI wheel, upgrades managed installations atomically, and refuses to overwrite or remove an unmanaged directory. Use genaug launch-skill status for a read-only integrity check and genaug launch-skill remove for rollback.

Planning writes the reviewable genaug-agent.yaml genaug/v2 Project contract. Existing genaug/v1 single-Agent contracts remain readable during the compatibility window. Inspection never writes or uploads application source. The beta contract enforces per-user memory, one disabled read-only capability, app-owned writes, no arbitrary code execution, and no automatic skill learning. Re-running inspection, planning, review, and provisioning is idempotent. Launch verification is contractual: every one of the 18 beta-required checks must be PASS. A required FAIL, SKIP, missing/stale receipt, plan-fingerprint mismatch, or unexecuted application command returns BLOCKED; optional warnings alone return READY_WITH_WARNINGS.

For an existing app, the friendliest first command is the guided wizard. It inspects the current app, detects frameworks, env files, OpenAI Responses call sites, prompts, tools, and webhooks, then writes a redacted .genaug/setup-plan.json without changing code or storing raw secrets.

genaug auth login
genaug init --interactive

genaug init --interactive runs the guided setup wizard without scaffolding a starter project. genaug setup is the explicit install/configure path with the same options.

genaug setup --capability code --capability browse --json
genaug setup --interactive
genaug setup --answers-template ./genaug-answers.template.json --json
genaug setup --guided --answers-file ./genaug-answers.json --json
genaug setup --guided --answers-file ./genaug-answers.json --project petstore-agent --configure-providers --json
genaug setup --bootstrap --project-name "Petstore Agent" --project-slug petstore-agent --print-env
genaug init --interactive --login --bootstrap --project-name "Petstore Agent" --project-slug petstore-agent --configure-providers --print-env
genaug setup --interactive --handoff-output .genaug/setup-handoff.md

Add --bootstrap after genaug auth login to create or select a project and mint a project runtime key through installer auth; the setup artifact stores only the masked key. Add --login to start browser installer auth inline before bootstrapping. Add --print-env to show the runtime env block once for your backend secret manager. Use --answers-template ... --json to generate the secret-free questionnaire before a human or coding agent fills it, then pass the completed file with --answers-file for non-interactive runs. --guided --json requires --answers-file so interactive prompts do not contaminate machine-readable output. Human guided setup writes .genaug/setup-handoff.md by default; use --handoff-output for a different path or --no-handoff to skip it.

Auth

genaug auth login
genaug auth whoami --json
genaug auth logout

genaug auth login starts browser installer auth by default: it opens the dashboard /cli/authorize approval page, creates an installer session for setup tasks, and keeps that session separate from runtime /v1/responses keys. The approval page identifies the requesting computer when the operating system provides a device name. After approval, the dashboard redirects to the CLI's loopback callback and the CLI exchanges the short-lived code automatically; if the callback cannot start or times out, it falls back to a paste-code prompt. genaug auth login --api-key ... remains available for operator/admin workflows and verifies the key against /api/v1/admin/me before writing local config. The CLI stores auth at ~/.genaug/config.yaml with owner-only file permissions; set GENAUG_CLI_CONFIG to use a custom path.

Environment overrides:

export GENAUG_ADMIN_API_KEY=gaadmlive...
export GENAUG_ADMIN_BASE_URL=https://api.generalaugment.com

GENAUG_ADMIN_API_KEY is management-only. GENAUG_API_KEY is application-runtime-only and never authorizes CLI control-plane operations. GENAUG_API_BASE_URL selects the API for both roles. Existing api_key config entries remain legacy management credentials; they are not silently upgraded or copied into the explicit runtime_api_key field.

Migrating an existing app

genaug migrate openai-responses --dry-run --json
genaug migrate openai-responses --apply --yes
genaug migrate openai-responses --apply --yes --branch genaug/openai-responses --push --create-pr

genaug migrate openai-responses inspects the app and generates a patch for OpenAI-compatible clients. It only edits files with --apply plus explicit confirmation or --yes. For coding-agent migration work, add --branch, --push, and --create-pr to create the branch, commit the safe edits, push, and open a GitHub pull request through the gh CLI.

Building a project from an OpenAPI spec

genaug integrate https://petstore3.swagger.io/api/v3/openapi.json
genaug integrate https://petstore3.swagger.io/api/v3/openapi.json --auto-deploy
genaug validate ./petstore-agent/genaug-agent.yaml

Use genaug init <name> for a starter agent before an OpenAPI spec exists, or genaug integrate <openapi-spec> --auto-deploy to create/update the project and register the generated OpenAPI tools in one pass. Deployment is performed by integrate --auto-deploy (there is no separate genaug deploy command). Without --auto-deploy, review the scaffold, then genaug validate ./<agent>/genaug-agent.yaml. The generated manifest is genaug-agent.yaml; both scaffolds include CODING_AGENT_PROMPT.md, the paste-ready backend handoff for a coding agent.

genaug init
genaug init dayplan-agent --tool web_search
genaug init --interactive

API keys

genaug keys create --project petstore-agent --name "Production backend"
genaug keys list --project petstore-agent --json
genaug keys update <key-id> --project petstore-agent --name "Renamed"
genaug keys revoke <key-id> --project petstore-agent

Providers and connectors

genaug providers setup --capability browse --project petstore-agent --api-key-env BROWSERBASE_API_KEY --health-check
genaug providers setup --capability video --json
genaug providers setup --provider codex-mcp --project petstore-agent --api-key-env OPENAI_API_KEY --health-check
genaug providers setup --provider fal --project petstore-agent --api-key-env FAL_API_KEY --health-check
genaug providers smoke --capability code --capability browse --capability video --json
genaug providers smoke --provider browserbase --project petstore-agent --api-key-env BROWSERBASE_API_KEY --json
genaug providers readiness --project petstore-agent --json
genaug connectors setup --name my-mcp --url https://example.com/mcp --health-check

providers setup reads the named env var once, stores the credential in General Augment custody, runs a health check, and writes only redacted provider setup evidence — raw secrets are never written to the setup artifact. providers smoke plans launch evidence per capability; add --evidence-output .genaug/provider-smoke-evidence.json to persist the redacted launch-smoke payload for launch reviews. providers readiness reports the provider readiness rows for a project. connectors setup writes MCP connector config through installer auth when --name plus --url or --command is supplied, and rejects raw API keys in MCP URLs.

For iMessage, use the npm helper on the Mac:

npx @general-augment/local-imessage setup --project dayplan-agent --write-prompt --write-config

Skills and tools

genaug skills design --job-type website-builder --project petstore-agent --apply
genaug skills list --project petstore-agent
genaug skills view <skill-name> --project petstore-agent
genaug skills apply ./SKILL.md --project petstore-agent
genaug skills delete <skill-name> --project petstore-agent

genaug tools catalog --project petstore-agent
genaug tools list --project petstore-agent --json
genaug tools toggle web_search --project petstore-agent --enable
genaug tools discovery --project petstore-agent
genaug tools explain-turn --project petstore-agent --requested-tool web_search --json
genaug tools add-mcp my-mcp --project petstore-agent --url https://example.com/mcp

Use exactly one transport per MCP server: --url for HTTP endpoints or --command for stdio servers.

Doctor, status, and smoke

genaug doctor
genaug doctor --project petstore-agent --user user_123 --json
genaug status --json
genaug smoke --idempotency-key smoke-replay-1 --metadata feature=spark
genaug smoke --project petstore-agent --structured --json
genaug smoke --project petstore-agent --memory-recall --import-evidence \
  --evidence-output .genaug/smoke-evidence.json --json

genaug doctor checks the resolved config path, base URL, API key presence, /health/ready, and /api/v1/admin/me without printing secret values. Add --project to verify read-only agent-cloud readiness (runtime policy, tool catalog, approvals, run timelines) and --user to include a tenant memory profile reachability check.

genaug smoke checks /health/ready and sends one project-keyed /v1/responses request using bearer auth. Use --idempotency-key, --request-id, --traceparent, and repeated --metadata key=value flags for replayable support/debug evidence. Use --project <slug> when the configured key is a management key and the app-facing request needs X-Project-ID. Use --structured for the default json_schema smoke response or --schema-file ./schema.json for an app-specific contract. Use --evidence-output to persist a redacted launch artifact (readiness result, response id, trace id, dashboard observability URL, secret-safety metadata). Add --memory-recall to prove /v1/responses recalls a seeded fact without prompt echo.

Verify

genaug verify --project petstore-agent

genaug verify runs project acceptance checks: project keys, hosted agent test, tools, logs, usage, usage limits, observability, runtime-policy model routing, memory lifecycle, and tool-call audit, then prints dashboard URLs for the same tenant.

Dashboard

genaug dashboard open --project petstore-agent

Local mock for offline tests

The deterministic local HTTP mock ships with the package and is launched as a module (it is not a genaug subcommand):

uv run --project packages/cli python -m platform_cli.local_mock \
  --host 127.0.0.1 --port 8787 --quiet

It serves offline app contract tests against /v1/responses, memory routes, project setup, OpenAPI tool registration, key management, logs, usage, observability, health checks, idempotency replays, trace metadata, structured-output fixtures, and semantic SSE fixtures. Point the SDKs at it with GENAUG_API_BASE_URL=http://127.0.0.1:8787.

Console entrypoint

The console command is defined in pyproject.toml:

[project.scripts]
genaug = "platform_cli.main:app"