uat: should investigate fix before asking severity

[skytect]

Human Note	During rapid:uat severity prompt, I am not sure what to click and just want to know "what would the fix look like?". Usually you don't just look at the problem and say its high severity, you look at the problem and the intended solution and say, maybe this fix is too complex to do now
Version	6.0.0
Skill	uat

Bug Description

When a UAT step fails, the skill immediately asks for severity (Critical/High/Medium/Low). But severity cannot be assessed in isolation from the fix. A "critical" failure with a one-line fix is different from a "medium" failure that requires rearchitecting a module.

The skill should proactively investigate the failure, analyze the relevant code, propose a likely fix, and estimate complexity BEFORE asking for severity. This gives the user the context needed to make an informed severity decision.

Steps to Reproduce

1. Run /rapid:uat <set-id> on a set with a failing criterion
2. A step fails (e.g., Biome doesn't flag a certain import pattern)
3. The skill immediately asks: "What severity level?" with Critical/High/Medium/Low
4. The user has no context about what the fix entails, how complex it is, or whether it's worth doing now

Root Cause / Suggested Fix

The UAT verification loop (Step 6 in SKILL.md) has a rigid flow: fail → severity → record → next step. There is no investigation step between failure detection and severity assignment.

Suggested fix: when a step fails, before the severity prompt, the skill should:

1. Read the relevant files for the failing criterion
2. Analyze what went wrong and what the fix would look like
3. Present a brief summary: what's broken, what the fix is, estimated complexity
4. Then prompt for severity with this context available

This way the user sees something like:

Step 9 FAILED: Biome didn't flag relative cross-package imports.

Fix analysis:
- File: biome.json
- Change: Add a second pattern to noRestrictedImports for relative paths
- Complexity: One-line config change
- Risk: Low

What severity level?

Workaround

User must independently investigate the failure before answering the severity prompt, or select "Other" to ask questions.

Co-Authored-By: Claude Opus 4.6 (1M context)