Google Cloud Tech posted a short but sharp thread on designing agent skills with SKILL.md. Their framing: thirty-plus tools now share the same YAML-and-spec layout, so format is no longer the interesting problem — content design is.
They call out five recurring patterns:
- Tool Wrapper — packages library-specific context dynamically
- Generator — produces consistent structured documents from templates
- Reviewer — scores code against severity-based checklists
- Inversion — the agent interviews the user before acting
- Pipeline — enforces sequential workflows with checkpoints
The money quote:
"Stop trying to cram complex and fragile instructions into a single system prompt. Break your workflows down, apply the right structural pattern, and build reliable agents."
That matches my own experience shipping LLM agents: the prompt is rarely the bottleneck; the missing piece is usually structural. Which of these five shapes fits the problem? Answer that, and the prompt almost writes itself.