AI Agent Stack Packs
AI makes unfamiliar stacks more reachable, but only if the agent loads the right docs, commands, traps, and quality gates. A stack pack is the durable unit that makes that happen.
Pack Contract
Every concrete stack pack should answer:
- When should you use this stack?
- When should you avoid it?
- Which official docs and examples should agents load?
- What is the smallest maintainable starter shape?
- Which commands create, run, test, lint, build, package, sign, and release?
- Which quality gates prove done?
- Which platform traps repeatedly break AI-generated work?
Minimum Files
Use Stack Capability Pack Template for the operational file set:
docs-index.mddecision-guide.mdstarter.mdcommands.mdquality-gates.mdpitfalls.mdagent-guardrails.mdreference-apps.md
Promotion Rule
Keep a stack pack as learn content until repeated usage proves that a live skill would save time and reduce mistakes. Do not promote every stack into an agent skill just because it exists.
Gotcha: A stack pack without commands and quality gates is not ready to drive implementation. It is research notes.