What problem does it solve?
It prevents costly mistakes and inconsistent workflows by standardizing how an agent decides between VS Code/MCP tools and terminal commands for reads, edits, searches, tests, and Git actions.
Core Features & Use Cases
- Tool-first routing rules: Directs the agent to use the specified tool for each task type, while avoiding unsafe terminal equivalents (e.g., no
cat/sed/awk/echo, no grep/find, no raw git CLI).
- Reliable test execution: Enforces running tests via the
runTests tool only, with explicit prohibitions on terminal test commands like pytest, jest, or dotnet test.
- Controlled script execution: Requires showing the script and asking the user to enable snippet execution tools before running code snippets (no terminal workaround running runtimes).
- Verification after edits: Requires language-native lint/type checks after modifications, since editor Problems may miss issues.
- Language-aware guidance via references: Uses dedicated references for Python, TypeScript, Java, and C# to choose the right checks and integrations.
Quick Start
Ask an AI to follow tool-usage rules for your request and it will select the correct tool for file edits and verification, run tests through runTests, and apply language-native lint/type gates using the appropriate reference guidance.