Lobster

Lobster runs multi-step tool pipelines as one repeatable tool call, with explicit approval/input checkpoints and resume tokens. Checkpoints belong to the Lobster runner, not a separate orchestration registry.

Why

Without Lobster, a multi-step job means many round-trip tool calls, with the model orchestrating every step. Lobster moves that orchestration into a typed runtime:

• One call instead of many: a single Lobster tool call returns a structured result for the whole pipeline.
• Approvals built in: side effects (send, post, delete) halt the workflow until explicitly approved.
• Resumable: a halted workflow returns a token; approve and resume without re-running earlier steps.

Lobster is a small, constrained DSL rather than a general scripting language: approve/resume is a durable, built-in primitive; pipelines are data (easy to log, diff, replay, review); the tiny grammar limits "creative" code paths so validation stays realistic; timeouts, output caps, sandbox checks, and allowlists are enforced by the runtime, not by each script. Each step can still call any CLI or script - generate .lobster files from other tooling if you want a richer authoring language.

Without Lobster, a recurring email triage looks like:

User: "Check my email and draft replies"→ openclaw calls gmail.list→ LLM summarizes→ User: "draft replies to #2 and #5"→ LLM drafts→ User: "send #2"→ openclaw calls gmail.send(repeat daily, no memory of what was triaged)

With Lobster, the same job is one call that halts for approval and resumes:

{ "action": "run", "pipeline": "email.triage --limit 20", "timeoutMs": 30000 }

{  "ok": true,  "status": "needs_approval",  "output": [{ "summary": "5 need replies, 2 need action" }],  "requiresApproval": {    "type": "approval_request",    "prompt": "Send 2 draft replies?",    "items": [],    "resumeToken": "..."  }}

How it works

The separately installed official @openclaw/lobster plugin runs Lobster workflows in-process using its embedded @clawdbot/lobster runtime. No external lobster subprocess is spawned; the tool call returns a JSON envelope directly. If the pipeline halts for approval or input, Lobster saves its continuation and returns a resume token. Approval requests can also carry a short approval ID. The call ends at the checkpoint; no process waits for the user's answer.

Enable

Lobster is an optional plugin tool, not installed or enabled by default. Install the official plugin:

openclaw plugins install @openclaw/lobster

Installation applies to a running Gateway automatically; otherwise it takes effect on the next startup. See Apply changes and inspect.

Then allow the tool globally:

{  "tools": {    "alsoAllow": ["lobster"]  }}

Or per-agent:

{  "agents": {    "entries": {      "main": {        "tools": {          "alsoAllow": ["lobster"]        }      }    }  }}

The tool is disabled entirely for sandboxed tool contexts.

If you need the standalone Lobster CLI for development or external pipelines (outside the embedded gateway runner), install it from the Lobster repo and put lobster on PATH.

Pattern: small CLI + JSON pipes + approvals

Build tiny commands that speak JSON, then chain them into one Lobster call. (Example command names below - swap in your own.)

inbox list --jsoninbox categorize --jsoninbox apply --json

{  "action": "run",  "pipeline": "exec --json --shell 'inbox list --json' | exec --stdin json --shell 'inbox categorize --json' | exec --stdin json --shell 'inbox apply --json' | approve --preview-from-stdin --limit 5 --prompt 'Apply changes?'",  "timeoutMs": 30000}

If the pipeline requests approval, resume with the token:

{  "action": "resume",  "token": "<resumeToken>",  "approve": true}

Example: map input items into tool calls:

gog.gmail.search --query 'newer_than:1d' \  | openclaw.invoke --tool message --action send --each --item-key message --args-json '{"provider":"telegram","to":"..."}'

JSON-only LLM steps (llm-task)

For a structured LLM step inside a workflow, enable the optional llm-task plugin tool and call it from Lobster:

{  "plugins": {    "entries": {      "llm-task": { "enabled": true }    }  },  "agents": {    "entries": {      "main": {        "tools": { "alsoAllow": ["llm-task"] }      }    }  }}

Important limitation: embedded Lobster vs openclaw.invoke

The installed Lobster plugin runs workflows in-process inside the gateway. In that embedded mode, openclaw.invoke does not automatically inherit a gateway URL/auth context for nested OpenClaw CLI tool calls.

That means this pattern is not currently reliable in the embedded runner:

openclaw.invoke --tool llm-task --action json --args-json '{ ... }'

Use the example below only when running the standalone Lobster CLI in an environment where openclaw.invoke is already configured with the correct gateway/auth context.

For openclaw.invoke and clawd.invoke, ambient OPENCLAW_TOKEN or CLAWD_TOKEN credentials are accepted only for localhost, 127.0.0.1, or [::1] destinations. To send credentials to another HTTP(S) endpoint, pass --token explicitly. This rule also applies to embedded workflows that explicitly configure a remote connection. This command argument is the remote Gateway credential, not the Lobster tool's approval-resume token parameter. If an invocation times out or fails after dispatch, Lobster does not retry it automatically, because the Gateway may already have performed the action.

openclaw.invoke --tool llm-task --action json --args-json '{  "prompt": "Given the input email, return intent and draft.",  "thinking": "low",  "input": { "subject": "Hello", "body": "Can you help?" },  "schema": {    "type": "object",    "properties": {      "intent": { "type": "string" },      "draft": { "type": "string" }    },    "required": ["intent", "draft"],    "additionalProperties": false  }}'

If you are using the embedded Lobster plugin today, prefer either:

• a direct llm-task tool call outside Lobster, or
• non-openclaw.invoke steps inside the Lobster pipeline until a supported embedded bridge is added.

See LLM Task for details and configuration options.

Workflow files (.lobster)

Lobster can run YAML/JSON workflow files with name, args, steps, env, condition, and approval fields. Set pipeline to the file path in the tool call.

name: inbox-triageargs:  tag:    default: "family"steps:  - id: collect    command: inbox list --json  - id: categorize    command: inbox categorize --json    stdin: $collect.stdout  - id: approve    command: inbox apply --approve    stdin: $categorize.stdout    approval: required  - id: execute    command: inbox apply --execute    stdin: $categorize.stdout    condition: $approve.approved

Notes:

• stdin: $step.stdout and stdin: $step.json pass a prior step's output.
• condition (or when) can gate steps on $step.approved.

Injected environment variables

Every step shell inherits the parent environment plus these Lobster-injected variables, so commands can reference resolved workflow args without embedding raw values into the command string:

• LOBSTER_ARG_<NAME> - one per workflow arg. The name is uppercased with each run of non-alphanumeric characters collapsed to _, so arg user-id becomes LOBSTER_ARG_USER_ID.
• LOBSTER_ARGS_JSON - every resolved arg as a single JSON string.

That is the complete injected set. There are no per-step output variables such as LOBSTER_STEP_<id>_STDOUT or LOBSTER_STEP_<id>_JSON_<field>; shells treat those names as unset, so parameter-expansion defaults can hide the error. Read a prior step's output through step references instead - $step.stdout, $step.json, or $step.json.<field> - in a stdin:, env:, or condition: value. (LOBSTER_STATE_DIR is a separate runtime setting for the state directory, not a per-run arg.)

Tool parameters

run

{  "action": "run",  "pipeline": "gog.gmail.search --query 'newer_than:1d' | email.triage",  "cwd": "workspace",  "timeoutMs": 30000,  "maxStdoutBytes": 512000}

Run a workflow file with args:

{  "action": "run",  "pipeline": "/path/to/inbox-triage.lobster",  "argsJson": "{\"tag\":\"family\"}"}

resume

{  "action": "resume",  "token": "<resumeToken>",  "approve": true}

For approvals, use token or approvalId from requiresApproval and a boolean approve. For input, use token from requiresInput and responseJson. To cancel either kind of checkpoint, use cancel: true instead of a decision. Supply exactly one of approve, responseJson, or cancel: true.

Structured input

A workflow input step or an inline ask stage returns needs_input with the question, a JSON Schema and a resume token. Optional defaults and subject provide suggested values and material to review. For example:

{  "status": "needs_input",  "requiresInput": {    "type": "input_request",    "prompt": "What feedback should be included?",    "responseSchema": { "type": "string" },    "resumeToken": "<resumeToken>"  }}

The agent presents the question in chat, then sends the user's answer as JSON:

{  "action": "resume",  "token": "<resumeToken>",  "responseJson": "\"Please shorten the introduction.\""}

responseJson can encode any value allowed by the returned schema, not just an object. Lobster validates the answer before continuing. Invalid JSON or an answer that does not match the schema leaves the checkpoint available for correction. A resume can return another question or approval request.

This is a chat/tool interaction, not an Inbox card or form. The plugin does not list pending checkpoints; retain the returned token to resume later. As with approval tokens, possession of an input token permits resume by a caller allowed to use the tool; tokens are not bound to an OpenClaw user or session.

Output envelope

Lobster returns a JSON envelope with one of four statuses:

• ok - finished successfully
• needs_approval - paused; requiresApproval carries a resumeToken and a short approvalId, either of which can resume the run
• needs_input - paused; requiresInput carries the question, answer schema and resumeToken
• cancelled - explicitly denied or cancelled

The tool surfaces the envelope in both content (pretty JSON) and details (raw object).

Approvals

If requiresApproval is present, inspect the prompt and decide:

• approve: true - resume and continue side effects
• approve: false - cancel and finalize the workflow

Use approve --preview-from-stdin --limit N to attach a JSON preview to approval requests without custom jq/heredoc glue. Resume state is stored as small JSON files under the Lobster state directory (~/.lobster/state by default, override with LOBSTER_STATE_DIR); the token itself only encodes a pointer to that state, not the full pipeline state.

Safety

• Local in-process only - workflows execute inside the gateway process; no network calls from the plugin itself.
• No secrets - Lobster doesn't manage OAuth; it calls OpenClaw tools that do.
• Sandbox-aware - disabled when the tool context is sandboxed.
• Hardened - timeouts and output caps enforced by the embedded runner.

Troubleshooting

Learn more

• Plugins
• Plugin tool authoring

Case study: community workflows

One public example: a "second brain" CLI + Lobster pipelines that manage three Markdown vaults (personal, partner, shared). The CLI emits JSON for stats, inbox listings, and stale scans; Lobster chains those commands into workflows like weekly-review, inbox-triage, memory-consolidation, and shared-task-sync, each with approval gates. AI handles judgment (categorization) when available and falls back to fixed rules when not.

• Thread: https://x.com/plattenschieber/status/2014508656335770033
• Repo: https://github.com/bloomedai/brain-cli

Related

• Automation - all automation mechanisms
• Tools Overview - all available agent tools
• Lobster plugin reference - manifest, config, and tool reference for the plugin