apply-skill

name	apply-skill
description	Loads an agentskills.io skill and executes its steps in an isolated ACA Sandbox. Use when the user asks to "apply", "run", "execute", "set up", "scaffold", "create a project using", or wants to follow a skill's instructions with actual command execution. Creates an isolated environment, runs the commands, and returns results.
metadata	{"author":"Troy Taylor","version":"1.0","pattern":"mutation"}

Apply a Skill (Execute in Sandbox)

What This Skill Does

Loads an agentskills.io skill, extracts its executable steps, and runs them in an isolated Azure Container Apps Sandbox. The sandbox provides a fresh Linux environment with the tools needed (dotnet, node, python, etc.). Commands run sequentially; output is returned to the user.

When to Activate

• User says "apply that skill" or "run that skill"
• User asks to "set up a project" or "scaffold" something
• User says "execute the steps" from a skill
• User wants to follow a skill's workflow with real execution
• User asks to "try it" after seeing skill guidance

Workflow

1. Load the skill. Use get_skill to fetch the full SKILL.md:

   get_skill(owner: "dotnet", repo: "skills", skill_path: "plugins/dotnet-ai/skills/technology-selection")

2. Parse executable steps. Read the skill content and identify:

   • Shell commands (in code blocks marked as bash, shell, powershell)
   • Package installs (dotnet add package, npm install, pip install)
   • File creation steps
   • Configuration commands

3. Choose the right sandbox image. Based on the skill's requirements:

   • .NET skills → image: ubuntu (has dotnet SDK)
   • Node.js skills → image: ubuntu (has node)
   • Python skills → image: ubuntu (has python)
   • General → image: ubuntu

4. Confirm with the user. Before executing, show what will be run:

   "I'll execute these commands in an isolated sandbox:

   1. dotnet new webapi -n MyApi
   2. cd MyApi && dotnet add package Microsoft.Extensions.AI
   3. dotnet build

   Proceed?"

5. Execute commands. Use execute_in_sandbox:

   execute_in_sandbox(commands: ["dotnet new webapi -n MyApi", "cd MyApi && dotnet add package Microsoft.Extensions.AI", "cd MyApi && dotnet build"], image: "ubuntu")

6. Report results. For each command, show:

   • Whether it succeeded (exit code 0) or failed
   • Relevant stdout output (trimmed to key information)
   • Any errors with suggestions

7. Offer follow-up actions:

   • "Want me to run more commands in the same sandbox?"
   • "Should I check the generated files?"
   • "Want me to clean up the sandbox?"

Output Format

Execution results

Applying skill: technology-selection

#	Command	Status	Output
1	dotnet new webapi -n MyApi	✓	Project created
2	dotnet add package Microsoft.Extensions.AI	✓	Package added (v9.1.0)
3	dotnet build	✓	Build succeeded, 0 warnings

Sandbox ID: sb-a1b2c3 (active for 10 more minutes)

Want me to:

• Run additional commands in this sandbox?
• Show the project structure?
• Clean up the sandbox?

On failure

#	Command	Status	Output
1	dotnet new webapi -n MyApi	✓	Project created
2	dotnet add package NonExistent.Package	✗	Package not found

Stopped at step 2. The package NonExistent.Package doesn't exist.

Suggestions:

• Check the package name on nuget.org
• The skill may reference a preview package — try adding --prerelease

Safety Rules

• Always confirm before executing. Show the commands and ask for approval.
• Stop on first failure. Don't continue if a command fails (exit code ≠ 0).
• Sandbox is isolated. Nothing in the sandbox can affect the user's machine or other systems. It's a throwaway environment.
• Auto-cleanup. Sandboxes auto-delete after 10 minutes of idle time.
• No secrets. Never pass user credentials, tokens, or API keys to sandbox commands. If a skill requires auth, note this and let the user decide.