Skillrail: One versioned source for AI agent skills, synced to every tool

skillrail1 pts0 comments

GitHub - gbouziden/skillrail: One versioned source for AI agent skills, compiled to Claude Code, Cursor, Copilot, Windsurf, and AGENTS.md. Zero dependencies. · GitHub

/" data-turbo-transient="true" />

Skip to content

Search or jump to...

Search code, repositories, users, issues, pull requests...

-->

Search

Clear

Search syntax tips

Provide feedback

--><br>We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

-->

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

Sign in

/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up

Appearance settings

Resetting focus

You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

{{ message }}

gbouziden

skillrail

Public

Notifications<br>You must be signed in to change notification settings

Fork

Star

master

BranchesTags

Go to file

CodeOpen more actions menu

Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit

History<br>9 Commits<br>9 Commits

docs

docs

.gitignore

.gitignore

LICENSE

LICENSE

README.md

README.md

cli.js

cli.js

package.json

package.json

View all files

Repository files navigation

skillrail

One skill source. Every agent surface.

Your org writes the same agent instructions five times: a Claude Code skill, a Cursor rule, a Copilot instructions file, a Windsurf rule, and an AGENTS.md section. They drift immediately, nobody knows which version is current, and there's no owner, no versioning, and no review gate.

Anthropic's own enterprise guidance tells you to solve this yourself:

"Custom Skills do not sync across surfaces… If your organization deploys Skills across multiple surfaces, implement your own synchronization process."<br>"Maintain an internal registry for each Skill with: Purpose… Owner… Version… Dependencies… Evaluation status."<br>— Anthropic Agent Skills enterprise docs

skillrail is that missing piece: a canonical, versioned skill registry that compiles to every tool your teams actually use.

How it works

Skills live once, in the open Anthropic SKILL.md format, with org metadata:

skills/<br>deploy-checklist/<br>SKILL.md # frontmatter: name, description, version, owner, targets<br>reference.md # optional supporting files (copied for Claude)

npm i -g skillrail<br>skillrail init # set up skillrail.json + skills/ in a repo<br>skillrail add deploy-checklist # scaffold a skill<br>skillrail sync # compile + write to every configured target

sync generates, idempotently, with do-not-edit provenance headers:

Target<br>Output

claude<br>.claude/skills// (verbatim, supporting files included)

cursor<br>.cursor/rules/.mdc

copilot<br>.github/instructions/.instructions.md

windsurf<br>.windsurf/rules/.md

agentsmd<br>one managed block in AGENTS.md (Codex, Jules, Amp, …)

Per-skill targeting: targets: [claude, cursor] in frontmatter overrides the project default.

Share skills across the org

Any git repo is a registry — platform team publishes, product teams consume:

skillrail pull git@github.com:acme/agent-skills.git deploy-checklist<br>skillrail sync

The lockfile records origin + commit for every installed skill.

Governance built in

skillrail check # lint: description present & ≤1024 chars, owner set, semver,<br># body ≤500 lines — exit 1 on failure (CI gate)<br>skillrail status # exit 2 if any target is stale, missing, or DRIFTED<br># (drift = someone hand-edited a generated file; flagged before sync overwrites it)<br>skillrail list # inventory: version, owner, targets, origin

Drop check + status into CI and skill quality/consistency becomes a merge requirement, exactly the lifecycle Anthropic's enterprise docs prescribe — but automated.

Why this exists (the market gap, verified mid-2026)

Agent observability/eval is saturated (Braintrust $80M Series B; Langfuse acquired by ClickHouse; promptfoo acquired by OpenAI).

Memory, orchestration, and interop are absorbed by funded startups and hyperscalers (Mem0, Letta; A2A in Copilot Studio and Bedrock AgentCore).

But skill/instruction management across tools and teams has no incumbent : vendors each ship their own silo format, Anthropic explicitly punts cross-surface sync to customers, and 97% of orgs report they haven't scaled agents across the organization.

Roadmap (the paid layer)

The CLI stays MIT/open-source. The hosted product adds what teams pay for:

Hosted registry — private org registry with RBAC, review/approval flow on skill versions, and audit log.

Usage telemetry — which skills get invoked, by which agents, and which correlate with successful runs; kill dead skills with data.

API/claude.ai surface push — sync skills to Anthropic API workspaces and claude.ai, not just repos.

Skill eval harness — run...

skillrail skills skill claude agent sync

Related Articles