Your AI assistant and Enterprise Skills: who does what

Cursor, Copilot, Claude Code, Windsurf, and Junie write software; Enterprise Skills decides whether the result is safe to ship, and can prove it. Below: what each side does, with the commands and file paths the product renders.

Your AI assistant

Generates software

  • Writes code from your prompts, autocompletes, explains, refactors. Enterprise Skills adds no step to that loop.
  • Behaves differently per editor and per model. Each assistant has its own memory shape, context rules, and defaults.
  • Has no opinion about whether the change should ship. An assistant will write code that fails your security bar, because nothing tells it the bar exists.

Enterprise Skills

Makes it production-ready

  • Carries one canonical skill library into every editor's native rule system, so the review bar is the same bar no matter which assistant wrote the code.
  • Runs the audits as executable skills with recorded outputs, not as advice: security, API contracts, test gaps, dependencies, deploy readiness.
  • Gates releases in CI on evidence, and countersigns every decision into a log anyone can verify against a published key.

One library, five native surfaces

One install renders whichever of these formats your license tier includes: plain files in your repository, generated and kept current by the CLI. There is no separate plugin to maintain.

npm install -g enterprise-skills
enterprise-skills init      # renders your licensed editors' native formats

Cursor

session start

Skills and rules land in .cursor/skills/ and .cursor/rules/ and are discovered automatically. The rules run always-on, so governance holds even when nobody invokes anything.

The command shown is the shipped first step from our getting-started path: it loads the authority pack, git status, and pending work.

VS Code + Copilot

@workspace

Copilot reads the same skill files through its instructions format plus AGENTS.md. Reference them in Copilot Chat with the workspace context.

Prefix a Copilot Chat request with @workspace so it reads the rendered skill files.

Claude Code

/security-audit

Claude Code loads CLAUDE.md governance and exposes each skill as a slash command.

Any installed skill runs by name; the security audit is the one shown.

Windsurf

.windsurfrules

Cascade rules render to .windsurfrules with skills under .windsurf/, including persistent memories.

The file shown is the rendered artifact; Windsurf loads it on its own.

JetBrains (Junie)

.idea/enterprise-skills/

The CLI renders skill files, context XML, and prompt templates into .idea/enterprise-skills/, from the same canonical library as the other surfaces. Junie's prompt library imports them manually.

The directory shown is the rendered artifact; import the prompts into Junie once per project.

The part no assistant does

Review inside the editor is advice; the gate result is enforced. In CI, one command produces a deterministic verdict across the quality domains your change touches, and a probe can check the deployed configuration rather than what the documentation claims about it.

enterprise-skills govern    # deterministic PASS/FAIL on the diff; --post completes the PR check
enterprise-skills probe run # checks the DEPLOYED config, not the docs

Every verdict is countersigned into a transparency log. Paste any decision record into the in-browser verifier.

What Enterprise Skills does not do

  • It does not write your features. Code generation stays entirely with your assistant.
  • It does not replace your CI. The gate runs inside the CI you already have and adds a verdict, not a pipeline.
  • It does not grade on trust. A domain nothing verified reports as unverified, never as passed.
Set it up in your editor