Date Compiled: 2026-04-28
Type: concept — factory-methodology
Related Questions: pipeline-design, resumable-workflows, human-in-the-loop-approvals, cron-automation, enterprise-operations
Related Articles: meta-crons, superada-enterprise-operations, superada-multi-agent-architecture, design-gate, full-pipeline
Source Attribution: compiled/sources/superada-enterprise-operations.md · raw/superada/blog-managing-crons.md


Lobster Pipelines — Typed JSON Envelope with Resumable Approvals

What Are Lobster Pipelines?

Lobster is the Enterprise Crew's workflow runtime for complex multi-step autonomous tasks. It consists of:

  1. A typed JSON envelope — a structured schema that wraps each step in a pipeline, making the state machine explicit and inspectable
  2. A resumable approvals workflow — human approval gates that can be paused, resumed, and do not require the human to be present at the exact moment the gate is reached

Complex cron jobs (those with 3+ deterministic sequential steps) are candidates for Lobster migration. The meta-crons system's Cron Watcher identifies these as LOBSTER_CANDIDATE tags, and the Auto-Lobster Converter validates them on Sundays.

Why Lobster Exists

Before Lobster, complex multi-step crons had a failure mode: if the human needed to approve step 3 but wasn't available, the cron would fail or stall — even if steps 1 and 2 had completed successfully. The entire pipeline had to restart when the human became available.

Lobster solves this by making the pipeline stateful and resumable:
- Each step writes its output to the envelope
- Approval gates are explicit states, not blocking points
- When the human approves, the pipeline resumes from the last completed step — no re-execution of already-completed work

The Envelope Schema

A Lobster pipeline is a JSON file with typed step definitions:

{
  "pipeline": "email-triage",
  "steps": [
    { "id": 1, "action": "fetch", "status": "done", "output": {...}, "approvedBy": "ada" },
    { "id": 2, "action": "classify", "status": "done", "output": {...}, "approvedBy": "henry" },
    { "id": 3, "action": "route", "status": "pending", "requiresApproval": "henry" }
  ]
}
  • Each step has a status: pending, in_progress, done, or blocked
  • Approval requirements are explicit per step
  • The envelope persists across sessions — a pipeline can be resumed days later

Active Lobster Pipelines

Nine Lobster pipelines are currently active in the Enterprise Crew:

Pipeline Purpose
hourly-ops.lobster Recurring operational tasks, hourly cadence
email-triage.lobster Inbound email classification and routing
entity-backup.lobster Scheduled data persistence
daily-ops.lobster Daily operational routine
secondary-inbox-monitor.lobster Secondary channel monitoring
morning-actions.lobster 8am UTC morning task batch
evening-data-collection.lobster End-of-day data aggregation
morning-data-pipeline.lobster Early morning data processing
production-deploy.lobster Production deployment with human approval gate

The production-deploy.lobster pipeline is the highest-stakes example: a multi-step build → test → approve → deploy pipeline where the human approval gate is intentional and mandatory, not a stall point.

How Crons Become Lobster Candidates

The meta-crons system's Cron Watcher reads cron payloads on every run. When it detects a cron with 3+ deterministic sequential steps, it tags that cron as LOBSTER_CANDIDATE.

The Auto-Lobster Converter (Sundays 2am UTC) then:
1. Scans all LOBSTER_CANDIDATE tagged crons
2. Runs lobster-converter-core.sh validation against each
3. Posts a dry-run summary to #upgrades for human review

No automatic migration — the conversion is validated and then approved by a human before it runs.

Connection to Design Gates

The approval-gate pattern in Lobster is a concrete implementation of design-gate methodology at the workflow level. Each human approval in a Lobster pipeline is a design gate — a quality threshold that work must pass before advancing to the next phase.

Connection to Factory Methodology

Lobster pipelines embody the principle that complex workflows need explicit state machines, not implicit sequencing. A cron with 5 sequential steps is not 5 crons — it is one stateful workflow, and the envelope makes that explicit.

This is adjacent to the full-pipeline concept in dark-factory-kb, which describes multi-stage build pipelines with gate-based advancement.


Sources: superada-enterprise-operations.md (compiled/sources/), blog-managing-crons.md (raw/superada/)

design-gate, full-pipeline, meta-crons, superada-enterprise-operations, superada-multi-agent-architecture