@agentled/cli/llms.txt

CLI for Agentled — manage workflows, apps, and knowledge from the command line. Zero context-window cost for AI agents.

# Agentled

> Workflow automation platform. Build, run, and manage agentic workflows via CLI or API. 100+ app integrations, AI steps, knowledge base, and credit-based execution.

## Quick Start

One command: browser sign-in → workspace folder → MCP auto-config → skill install → knowledge probe.

```bash
npx @agentled/cli setup
```

When an AI agent knows which client it is configuring, pass the target explicitly
instead of relying on auto-detection:

```bash
npx @agentled/cli setup --mcp-client codex
npx @agentled/cli setup --mcp-client claude-code
npx @agentled/cli setup --mcp-client cursor
npx @agentled/cli setup --mcp-client windsurf
npx @agentled/cli setup --mcp-client claude-desktop
```

OpenClaw and Hermes are skill-install targets, not MCP auto-config targets yet:

```bash
npx @agentled/cli auth login
npx @agentled/cli skills install --target openclaw
npx @agentled/cli skills install --target hermes
```

Configure MCP for OpenClaw/Hermes in the client's native MCP settings using the
saved Agentled credentials or an `AGENTLED_API_KEY`.

For workspace switching or failed browser auth recovery, re-run with
`--reauth`, for example `npx @agentled/cli setup --reauth --mcp-client codex`.
Restart/reconnect the AI client after setup so MCP picks up the new credentials.

Global home is `~/.agentled/` (config + on-demand examples). Per-workspace artifacts live in `agentled_<slug>/` in the current directory. Best-practice patterns: https://github.com/Agentled/agentic-ops.

## Components (à-la-carte)

```bash
npm i -g @agentled/cli
agentled auth login            # browser auth + workspace selection + skill install
agentled init                   # scaffold agentled_<slug>/
agentled auth current
agentled workflows list
agentled workspace set-output-pin <workflow-id> <output-page-pathname> --label "Report"
```

The CLI can store multiple workspace profiles. Use `agentled auth list`,
`agentled auth use <workspace>`, or `agentled --workspace <workspace> ...`
to switch targets.

## Docs

- [Full Reference](llms-full.txt): Complete single-file reference for LLM consumption — all CLI commands, workflow schema, app actions, and examples
- [Design Principles](patterns/v1/00-design-principles.md): The four-pillar authoring contract — idempotency, small scope, KG-list-as-connector, design-as-data — read this **before authoring any workflow**.

## Scaffolds (zero-iteration starters)

Every scaffold is preflight-clean. List them with `agentled workflows scaffold --list`; copy one with `agentled workflows scaffold <slug> --out my.json`.

Idempotent sourcing + processing (Pattern 15 — the user's default for "find leads then process on a schedule"):

- `source-to-kg` — schedule → generate queries → extract → `kg.upsert-rows` with `userKey` + `status:new`. Re-runs dedup via userKey.
- `kg-process-update` — schedule → `kg.read-list(status:new)` → loop → `kg.update-rows(status:processed)`. Pair with `source-to-kg`.

Composition:

- `child-with-return` — child workflow with `internal:true` + `return` step. Called via `agentled.call-workflow`.
- `orchestrator-kg-loop` — orchestrator that loops `call-workflow` over a KG list, waits for `loop_completion`, persists results.

See `scaffolds/README.md` for the full catalog.