Work Units
Atomic units of work with YAML specs, worktree isolation, and lifecycle tracking. Learn more →
The Software Delivery Pack is the default domain pack shipped with LumenFlow. It provides everything you need for structured software delivery — from work unit lifecycle to quality gates to agent coordination.
Work Units
Atomic units of work with YAML specs, worktree isolation, and lifecycle tracking. Learn more →
Lanes
Scoped work streams with WIP limits, lock policies, and path-based access control. Learn more →
Gates
Config-driven quality checks (format, lint, typecheck, test) enforced as kernel policies. Learn more →
Memory Layer
Session tracking, checkpoints, coordination signals, and context recovery for agents. Learn more →
The pack registers 100+ CLI commands across these categories:
| Category | Examples | Count |
|---|---|---|
| WU Lifecycle | wu:create, wu:claim, wu:prep, wu:done | 20+ |
| Memory & Sessions | mem:checkpoint, mem:inbox, mem:signal | 14 |
| Initiatives | initiative:create, initiative:status | 8 |
| Gates & Validation | gates, wu:validate, lumenflow-doctor | 6 |
| Packs | pack:author, pack:install, pack:validate | 6 |
| Metrics & Flow | flow:report, flow:bottlenecks, metrics | 5 |
| Agent & Orchestration | wu:brief, wu:delegate, orchestrate:monitor | 7 |
See the CLI Reference for the complete command list.
The recommended path for new packs is template-first authoring:
Template generation enforces least-privilege defaults and validation catches unsafe scope/network configurations before install or publish.
Install command shape:
The Software Delivery Pack ships with policies that the kernel evaluates automatically:
on_completion with deny-wins semanticsSee the Policy Engine for how these are evaluated.
If you are a software team using AI agents, the Software Delivery Pack replaces ad-hoc methodologies with a structured, evidence-based workflow. Here is how it compares to traditional approaches:
Traditional tickets are vague — “Add user authentication.” Work Units are executable specifications with acceptance criteria, code paths, and test paths. An AI agent can read a WU spec and implement it autonomously. A human can verify completion by checking the criteria.
Scrum batches work into time-boxed sprints. The Software Delivery Pack flows work through parallel lanes — each with a WIP limit (typically 1 for solo work, 2 for teams). Work flows continuously instead of waiting for sprint boundaries.
Code review is a bottleneck. The Software Delivery Pack replaces manual review with automated quality gates — format, lint, typecheck, and test checks that must all pass before work can be completed. Critical changes can still require human approval via escalation triggers.
Daily standups exist because humans forget context. The Software Delivery Pack’s memory layer provides checkpointing and context recovery, so both humans and AI agents can resume exactly where they left off.
Kanban boards have “Done” columns. The Software Delivery Pack creates immutable stamps — created only when all acceptance criteria pass, all gates pass, and work is merged. No ambiguity.
| Aspect | Scrum | Kanban | Shape Up | Software Delivery Pack |
|---|---|---|---|---|
| Time unit | 2-week sprints | Continuous | 6-week cycles | Continuous |
| Work item | User story | Card | Pitch | Work Unit (WU) |
| Progress | Story points | Cycle time | Hill charts | Gates + stamps |
| Parallelism | Team velocity | WIP limits | Team per pitch | Lane-based WIP |
| Reviews | Sprint review + PR | PR review | Demo | Automated gates |
| AI-ready | No | Partially | No | Yes |
| Context | Human memory | Board state | Write-up | Memory layer |