Getting started¶
Skills are markdown instructions the agent reads when your task matches their description. They keep behavior consistent: when to align, when to test, when to file issues, and what not to do.
Where skills live¶
- In Cursor:
.cursor/skills/<name>/SKILL.md,skills/<name>/, or paths you configure. - On this site: Browse the Skills catalog—each skill has its own page under Skills in the sidebar.
How to invoke a skill¶
Use an @mention in chat so the agent loads the right skill before writing code or tests.
Minimal pattern
The agent should confirm the behavior list, then write one failing test, then minimal code.
Choose a workflow¶
flowchart LR
A[New idea / PRD] --> B{Artifact?}
B -->|Local plan| C[prd-to-plan]
B -->|GitHub issues| D[to-issues]
A --> E[grill-me]
F[Unfamiliar repo] --> G[inherited-codebase-onboarding]
H[Bug] --> I[triage-issue]
J[Implement slice] --> K[aligned-implementation + tdd]
- Product clarity: grill-me, prd-critique
- Plan or tickets: prd-to-plan, to-issues
- Orientation: inherited-codebase-onboarding
- Shipping a slice: aligned-implementation + tdd
- GitHub: issue-to-pr after your branch is ready
Rules that work well¶
- One anchor per thread — issue number, plan section, or pasted scope block; avoid mixing unrelated features.
- Approve before RED — for
aligned-implementationandtdd, settle the public API and “done for this thread” before the first failing test. - Vertical slices — each PR or thread should deliver something end-to-end and demoable, not “all models then all controllers.”
- Escalate with the right skill — tangled code → improve-codebase-architecture or request-refactor-plan; Neon egress → neon-postgres-egress-optimizer.
Next steps¶
- Open the Skills catalog and jump to the pages you reuse most.
- Walk the calendar refresh recipe to see skills in sequence.
- Copy from Cookbook: quick examples.