fix-bug

name	fix-bug
version	0.1.0
description	Handles bug-fix investigation and testing when start-task routes a Bug category task. Orchestrates root cause investigation (DIAGNOSE_AND_FIX_BUG capability) and a TDD hard gate to prevent recurrence (APPLY_TDD capability), resolved via the runtime-bindings table. Returns to start-task for QA verification and shipping. Always called from start-task; never invoked directly.

Fix Bug

Focused orchestrator for bug investigation and fix. Contains no debugging or testing logic — sequences two capabilities with explicit gates, then returns to start-task for QA verification and shipping. Always called from start-task, never directly. The concrete skill behind each capability is resolved per runtime from plugins/cortex-workflow/references/runtime-bindings.md.

Inputs

Receives full task context from start-task:

• Task name, notes, custom fields (category, task ID, sprint)
• Subtasks (incomplete = remaining work, completed = done)
• Comments and attachments
• QA investigation report (from start-task's QA: Investigate Bug)
• Branch name (already created and checked out)
• Task URL

Step 1: Root Cause Investigation

Invoke the DIAGNOSE_AND_FIX_BUG binding with the full task context as the bug report. If the bound skill is unavailable, stop and tell the user to install its plugin alongside cortex-workflow before continuing (see references/skill-dependencies.md in start-task for recovery commands).

Do not proceed to Step 2 until the investigation confirms both:

• Root cause identified with specificity (not "it was broken")
• Fix implemented on the current branch

Step 2: TDD Hard Gate

This step cannot be skipped under any circumstance.

After the investigation declares the fix ready, invoke the APPLY_TDD binding with these explicit requirements:

1. Write a regression test that:

   • Would fail if the fix were reverted
   • Has a name that makes the bug recognizable (e.g., test_session_expiry_uses_utc_not_local)

2. Run the full test suite. All three must be true before completing:

   • The new regression test passes
   • No previously-passing tests are now failing
   • Any pre-existing failures are documented as pre-existing

If the gate cannot be satisfied (tests stay red, no meaningful test can be written):

• Do NOT return to start-task
• Restart the DIAGNOSE_AND_FIX_BUG binding from the beginning — the fix is not considered complete
• If the gate still cannot be satisfied after a second pass, halt and surface the blocker to the user with a summary of what was attempted

Output

When both steps pass, return to start-task with:

• Summary of root cause and fix (from the investigation)
• Regression test name and file path
• Test run output confirming all green

Start-task handles QA verification (QA: Verify Fix) and shipping (Step 12) from here.