Skip to content
| Marketplace
Sign in
Visual Studio Code>AI>Agent Experience GraphNew to Visual Studio Code? Get it now.
Agent Experience Graph

Agent Experience Graph

Agent Experience Graph

| (0) | Free
Discover, recommend, rate, and measure agent skills directly in VS Code.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Agent Experience Graph for VS Code

An early IDE integration for discovering, recommending, rating, and measuring agent skills directly inside a developer workspace.

Install

Install Agent Experience Graph from the Visual Studio Code Marketplace, then open a workspace containing SKILL.md or capability.json files.

For a local VSIX build:

npm install
npm run package
code --install-extension agent-experience-graph-0.1.0.vsix

MVP Features

  • Discovers local SKILL.md and capability.json files.
  • Recommends skills for selected code/text or a task description.
  • Opens the selected skill or copies its path for use by an agent.
  • Records local skill-use counts and estimated selection-context tokens.
  • Lets developers rate skills from 1 to 5.
  • Displays usage, estimated tokens, and average ratings.

Metrics remain local in .aeg/skill-metrics.json by default. This MVP does not send telemetry to a hosted service.

Run Locally

cd integrations/vscode
npm install
npm run compile

Open this directory in VS Code and press F5 to launch an Extension Development Host.

Commands

Open the Command Palette and run:

  • AEG: Discover Workspace Skills
  • AEG: Recommend Skill for Current Task
  • AEG: Rate a Skill
  • AEG: Show Skill Metrics

You can also select text in an editor, right-click, and choose AEG: Recommend Skill for Current Task.

Skill Discovery

The extension scans these workspace patterns by default:

[
  "**/SKILL.md",
  "**/capability.json"
]

They can be changed with aeg.skillGlobs.

Local Metrics Format

{
  "skills/example/SKILL.md": {
    "uses": 4,
    "totalEstimatedTokens": 820,
    "ratings": [5, 4],
    "lastUsed": "2026-07-15T12:00:00.000Z"
  }
}

The current token number is an estimate based on text length. A production version should collect actual model usage from supported agent runtimes.

Next Steps

  • Semantic embeddings and hybrid ranking instead of keyword overlap.
  • Parse richer skill manifests: inputs, outputs, tools, constraints, and compatibility.
  • Connect to Codex, OpenAI Agents SDK, Claude Code, MCP, and other runtimes.
  • Record actual tokens, cost, latency, retries, validation results, and success rate.
  • Add trusted automatic skill loading/install with permission boundaries.
  • Publish team and community skill leaderboards.
  • Package and publish to the VS Code Marketplace and Open VSX.
  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
© 2026 Microsoft