Your teams rewrite the same AI agent instructions five times — Claude skill, Cursor rule, Copilot instructions, Windsurf rule, AGENTS.md. They drift by Tuesday. skillrail gives skills a single versioned home and compiles them everywhere.
"Custom Skills do not sync across surfaces… implement your own synchronization process."
"Maintain an internal registry for each Skill with: Purpose… Owner… Version… Evaluation status." — Anthropic, Agent Skills enterprise documentation
Every agent vendor ships its own silo format and no cross-tool story. skillrail is the missing registry and sync layer — open source, zero dependencies, works with the formats you already have.
# skills/deploy-checklist/SKILL.md — open Anthropic format + org metadata --- name: deploy-checklist description: Follow the org deploy checklist before shipping to production. version: 2.1.0 owner: platform-team targets: all --- $ skillrail sync create .claude/skills/deploy-checklist/SKILL.md create .cursor/rules/deploy-checklist.mdc create .github/instructions/deploy-checklist.instructions.md update AGENTS.md (managed block) $ skillrail status drifted .cursor/rules/deploy-checklist.mdc — edited in place; port changes to the canonical skill
Platform team publishes one skills repo; every team pulls from it, commit-pinned in a lockfile.
Every skill carries an owner and semver. skillrail check fails CI when governance fields are missing.
skillrail status catches hand-edited generated files before a sync silently overwrites someone's work. Exit codes wired for CI.
Canonical format is Anthropic's open SKILL.md. Outputs are the plain files each tool already reads. Delete skillrail anytime; your files stay.
| Surface | Output |
|---|---|
| Claude Code | .claude/skills/<name>/ (verbatim, supporting files included) |
| Cursor | .cursor/rules/<name>.mdc |
| GitHub Copilot | .github/instructions/<name>.instructions.md |
| Windsurf | .windsurf/rules/<name>.md |
| Codex / Jules / Amp… | managed block in AGENTS.md |
The CLI is MIT forever. The hosted layer adds what platform teams pay for: a private registry with review & approval on skill versions, usage telemetry (which skills agents actually invoke — kill dead ones with data), push to Anthropic API / claude.ai surfaces, and a pre-promotion eval harness.