Software Factory gives an agent a fixed set of terms for designing a system of agents that turns tickets, PRs and schedules into merged code on remote infrastructure. It defines task, trigger, run, runner, ticket tracker, ticket source, git host, orchestrator and observability platform, and steers the conversation back to those terms.
The skill adds a design method and four surveys of products per slot, plus a catalogue of example tasks for common problems. It is user-invoked only.
What You Get
- A glossary of nine terms with the words to avoid for each
- A start-slow method built on automated checks, automated review and human review
- Surveys of ticket trackers, git hosts, runners and observability platforms
- A catalogue of example tasks: explore, implement, review, rework, decompose, ratchet, fix, architecture and retro
Common Use Cases
- Sketching a factory that turns labelled tickets into draft pull requests
- Choosing a runner type for review runs versus implementer runs
- Deciding which ticket tracker and git host fit a team that already uses Linear or GitHub
- Planning a daily ratchet task that migrates a codebase to a stricter lint rule
What Makes It Unique
Most agent-orchestration skills ship code that runs a pipeline. This one ships language and judgement: a controlled vocabulary, a rule to add one automated trigger at a time with throttles, and surveys that list the questions deciding each slot. It tells the agent to buy the generic services and build the tasks that carry your design.
Under The Hood
The skill is Markdown only. The main file holds the glossary and method, and it points to five reference files that the agent reads on demand: one survey per service slot and one file of example tasks. It runs no scripts and calls no services itself. The surveys carry a 2026-10 date and tell the agent to research current products itself, since prices and betas change.