Add structured convergence summary to audit output

[mickeylorenzini]

Problem

Harden's Step 4 says "Stop. Deliver." but doesn't define what the final output looks like. The model infers inconsistently — sometimes a detailed summary, sometimes just "Done." This makes it impossible to:

• Verify harden did thorough work vs. gave up early
• Compare audit runs across different deliverables
• Feed structured output to an eval tracking tool for performance benchmarking

Proposed Solution

Add a mandatory output template to harden's Step 4. Every audit run ends with a consistent format:

Audit complete: [N] passes to convergence
Found: [N] 🔴, [N] 🟡, [N] 🟢
Fixed: [N] 🔴, [N] 🟡 — [summary of resolution]
Remaining: [N] 🟢 (below threshold)
Key changes: [1-2 sentence summary of most impactful fixes]

Why This Matters

This is the foundation for measurable harden performance. Without structured output, you can't compare runs, track regressions when adding new features, or benchmark whether a new audit dimension actually improved quality. Enables future eval tracking companion tool.

Likely Affected Files/Modules

• SKILL.md — Step 4 modification, output template definition

Acceptance Criteria

• Output template defined and added to Step 4 in SKILL.md
• Template includes: pass count, severity counts, fix summary, remaining items
• Format is parseable (consistent delimiters, predictable structure)
• Tested: run harden on a deliverable, verify output matches template
• Existing harden behavior unchanged except for final output format