investigate

name	investigate
description	Investigate a failing test by running it, emitting IR, and analyzing the error. Use when the user says "investigate", "investiga", "debug test", or wants to understand why a test fails.
user-invocable	true
argument-hint	<test-file-or-suite>

Delegation: Use the Agent tool to dispatch a deep-analysis-reviewer agent with model: opus for this task.

Investigation Workflow

1. Run the Failing Test

Use mcp__tml__test with:

• path or suite: from $ARGUMENTS
• verbose: true
• no_cache: true
• fail_fast: true (stop at first failure)

2. Capture Error

Parse the test output to identify:

• Which specific test function failed
• The error type (compilation, runtime assertion, codegen, linker)
• The error message and location

3. Deep Analysis

Based on error type:

Compilation error (T0xx, P0xx):

• Use mcp__tml__check to get type-check diagnostics
• Use mcp__tml__explain if there's an error code
• Read the source file around the error location

Runtime/assertion error:

• Read the test file to understand what it's testing
• Trace the assertion to the library implementation
• Read the library source to find the bug

Codegen/linker error:

• Use mcp__tml__emit-ir to see the generated IR
• Look for undefined symbols, type mismatches, or missing functions
• Check if it's a lazy-lib resolution issue

4. Report

Present:

• Error: What went wrong
• Root cause: Why it's happening
• Suggested fix: What to change and where
• Affected files: Which files need modification