github / github/awesome-copilot
`meta-agentic-project-scaffold` is a VS Code agent but ships to the Copilot CLI and desktop app
- Dominant language
- JavaScript
- Stars
- 39k
- Forks
- 5k
- Avg merge
- 1d 22h
- Merged PRs (30d)
- 113
Description
👋 Hi team,
Opening this because I noticed this while working with a customer on a related issue. The `awesome-copilot` plugin (v1.1.0) bundles one agent, `meta-agentic-project-scaffold`, whose content is written for VS Code.
- the tool list is entirely VS Code tool names (`vscodeAPI`, `openSimpleBrowser`, `runNotebooks`, `readCellOutput`, `terminalSelection`, `activePullRequest`, `copilotCodingAgent`)
- the prompt instructs the agent to return vscode-insiders install links
- it pins `model: "GPT-4.1"`, overriding the model the user selected
It is nonetheless packaged and surfaced as a Copilot agent. CI emits it to `com.github.copilot/agents/`, and installing the plugin puts "Meta Agentic Project Scaffold" in the agent picker of both the Copilot CLI (origin `plugin`) and the desktop app, where it can be selected and run.
Copilot App
Copilot CLI
Related: #610 flagged this same file for `activePullRequest`, `copilotCodingAgent`, and the renamed `changes` tool. It was closed as completed on 2026-03-09.
Maybe refresh the tool names, and add `target: vscode` to declare the intended environment? Worth flagging that `target` currently has no effect in the Copilot CLI or desktop app, so it will not remove the agent from those pickers on its own — I have filed that separately against `github/copilot-cli` in https://github.com/github/copilot-cli/issues/4806.
Contributor guide
Research direction
Start with the meta-agentic-project-scaffold agent content and the CI path that emits agents to com.github.copilot/agents/. Compare its VS Code-specific tools, prompt, and model setting with the packaging rules and related issue #610. Done should be defined as preventing this VS Code-only agent from being incorrectly surfaced in unsupported Copilot environments, with validation for the affected outputs.
Written by the indexing model from the issue text.
Assessment
- Domain
- ai, tooling
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 52/100