Reference
SKILL.md reference
Pages covering the parts of a SKILL.md that are easy to get subtly wrong, the limits on the plugin API, and the runtime SDK. Each has a plain-Markdown twin at the same URL with .md on the end.
- SKILL.md anatomyWhat each part of a SKILL.md file is for: the frontmatter, the body, and the references, scripts and assets folders beside it.
- SKILL.md frontmatter referenceEvery SKILL.md frontmatter field, field by field: name, description, allowed-tools, license, compatibility and metadata, with their constraints.
- allowed-tools in a SKILL.mdWhat allowed-tools controls, the token grammar it accepts, and why splitting the field on whitespace tears a scoped token like Bash(pnpm test:*) in two.
- Claude Code vs OpenClaw: what each host readsThe context files Claude Code and OpenClaw actually read, where skills land in each, and what that means for writing context that works in both.
- Keeping skills current across a teamHow to stop shared agent skills going stale: syncing from one source of truth, review before publish, and versioned releases you can roll back.
- Rate limitsThe request limit on the Contexory plugin API and sign-in pages: 300 requests per minute per IP address, and what to do when you get a 429.
- Contexory SDK: resolve released context at runtimeResolve an agent's released prompt, skills and tool descriptions from your own code, with a local cache, a last-good fallback and signed responses.