Lobster
Execute multi-step workflows with approval checkpoints. Lobster pipelines are deterministic (no LLM variance), resumable via tokens, and return structured JSON output.
When to Use Lobster
| User intent | Use Lobster? |
|---|---|
| "Triage my email" | Yes — multi-step, may send replies |
| "Send a message" | No — single action, use tool directly |
| "Check my email every morning and ask before replying" | Yes — scheduled workflow with approval |
| "What's the weather?" | No — simple query |
| "Monitor this PR and notify me of changes" | Yes — stateful, recurring |
Do not use for simple single-action requests, queries needing LLM interpretation mid-flow, or one-off non-repeatable tasks.
Workflow
- Identify whether the user request involves multi-step automation, approval gates, or recurring operations — if yes, use Lobster.
- Run the pipeline:
{ "action": "run", "pipeline": "gog.gmail.search --query 'newer_than:1d' --max 20 | email.triage" } - Check the response
statusfield:"ok"— pipeline completed; presentoutputto the user."needs_approval"— present therequiresApproval.promptand items to the user.
- If approval needed, collect user confirmation and resume:
{ "action": "resume", "token": "<resumeToken>", "approve": true } - Present final output to the user.
Key Behaviors
- Deterministic: Same input produces same output (no LLM variance in pipeline execution).
- Approval gates:
approvecommand halts execution, returns aresumeToken. - Resumable: Use
resumeaction with the token to continue after approval. - Structured output: Always returns JSON envelope with
protocolVersion,status,output, andrequiresApproval.
Example Pipelines
# Email triage — classify into needs_reply, needs_action, fyi
gog.gmail.search --query 'newer_than:1d' --max 20 | email.triage
# Email triage with approval gate
gog.gmail.search --query 'newer_than:1d' | email.triage | approve --prompt 'Process these?'
Guardrails
- Use local or trusted tools only; avoid untrusted install pipelines.
- Do not paste secrets/tokens into chat output.
- Keep outputs in deterministic local paths for reproducible review.