Audit Resolution

Purpose: Present audit discrepancies to user and capture decisions for each finding (HITL).

Use when: Investigation phase has completed and you have audit discrepancies that need user decisions before remediation.

Critical: This is a HITL skill - EVERY discrepancy requires a user decision. NO auto-fixes.

---

Workflow

1. Present Summary

   • Total files audited
   • Total discrepancies found
   • Breakdown by severity (critical, warning, info)

2. Sort Discrepancies

   • Order: critical → warning → info
   • Within severity: alphabetical by file path

3. For Each Discrepancy (Loop)

   • Show file path and line number
   • Show diff in table format:

     Aspect	Template	Actual
     field	expected	found

   • Show severity level
   • HITL: Ask user "What should I do?"

4. Present Decision Options

   [1] Apply template - Fix file to match template
   [2] Update template - Template is wrong, create PR to update it
   [3] Ignore - Accept this deviation
   [4] Custom - Provide specific instruction
   

5. Capture User Decision

   • Record: discrepancy ID, action choice, custom instruction (if [4])
   • Validate choice is 1-4
   • If [4], prompt for custom instruction

6. Record All Decisions

   • Build decisions array for remediation phase
   • Count decisions by type
   • Prepare summary for next phase

---

Input

From /skill audit-investigation:

{
  discrepancies: [
    {
      id: string,              // Unique identifier
      file: string,            // Absolute path
      line: number,            // Line number
      field: string,           // Config field name
      expected: string,        // Template value
      actual: string,          // Current value
      severity: "critical" | "warning" | "info",
      agent: string,           // Config agent that found it
      template_source: string  // Path to template file
    }
  ],
  summary: {
    files_audited: number,
    total_discrepancies: number,
    by_severity: {
      critical: number,
      warning: number,
      info: number
    }
  }
}

---

Output

{
  decisions: [
    {
      discrepancy_id: string,
      action: "apply" | "update_template" | "ignore" | "custom",
      custom_instruction?: string,  // Only if action === "custom"
      discrepancy: {                // Copy of original discrepancy
        file: string,
        line: number,
        field: string,
        expected: string,
        actual: string,
        severity: string,
        agent: string,
        template_source: string
      }
    }
  ],
  summary: {
    total_discrepancies: number,
    apply_count: number,
    update_template_count: number,
    ignore_count: number,
    custom_count: number
  }
}

---

Presentation Format

Summary Block

## Audit Results Summary

- Files audited: {files_audited}
- Total discrepancies: {total_discrepancies}
  - Critical: {critical_count}
  - Warning: {warning_count}
  - Info: {info_count}

I will now walk through each discrepancy for your decision.

Per-Discrepancy Block

---

### Discrepancy {n}/{total} [{severity}]

**File:** `{file}`
**Line:** {line}
**Field:** `{field}`
**Agent:** {agent}

**Difference:**
| Aspect | Template | Actual |
|--------|----------|--------|
| {field} | {expected} | {actual} |

**Template source:** `{template_source}`

**What should I do?**

1. Apply template - Fix file to match template
2. Update template - Template is wrong, create PR to update it
3. Ignore - Accept this deviation
4. Custom - Provide specific instruction

Your choice (1-4):

After All Decisions

## Resolution Summary

Decisions captured for {total_discrepancies} discrepancies:

- Apply template: {apply_count}
- Update template: {update_template_count}
- Ignore: {ignore_count}
- Custom: {custom_count}

Proceeding to remediation phase.

---

Validation Rules

1. Every discrepancy gets a decision - No skipping
2. Invalid choice → re-prompt - Must be 1-4
3. Custom requires instruction - If [4], must collect free-text
4. Sort by severity first - Critical issues presented first
5. Copy full discrepancy - Decisions include all context for remediation

---

Example

Input:

{
  "discrepancies": [
    {
      "id": "disc-001",
      "file": "/path/to/eslint.config.js",
      "line": 23,
      "field": "rules.no-console",
      "expected": "error",
      "actual": "warn",
      "severity": "warning",
      "agent": "eslint-agent",
      "template_source": "plugins/metasaver-core/skills/config/eslint/templates/base.js"
    }
  ],
  "summary": {
    "files_audited": 1,
    "total_discrepancies": 1,
    "by_severity": {
      "critical": 0,
      "warning": 1,
      "info": 0
    }
  }
}

Presentation:

## Audit Results Summary

- Files audited: 1
- Total discrepancies: 1
  - Critical: 0
  - Warning: 1
  - Info: 0

I will now walk through each discrepancy for your decision.

---

### Discrepancy 1/1 [WARNING]

**File:** `/path/to/eslint.config.js`
**Line:** 23
**Field:** `rules.no-console`
**Agent:** eslint-agent

**Difference:**
| Aspect | Template | Actual |
|--------|----------|--------|
| rules.no-console | error | warn |

**Template source:** `plugins/metasaver-core/skills/config/eslint/templates/base.js`

**What should I do?**

1. Apply template - Fix file to match template
2. Update template - Template is wrong, create PR to update it
3. Ignore - Accept this deviation
4. Custom - Provide specific instruction

Your choice (1-4):

User response: 1

Output:

{
  "decisions": [
    {
      "discrepancy_id": "disc-001",
      "action": "apply",
      "discrepancy": {
        "file": "/path/to/eslint.config.js",
        "line": 23,
        "field": "rules.no-console",
        "expected": "error",
        "actual": "warn",
        "severity": "warning",
        "agent": "eslint-agent",
        "template_source": "plugins/metasaver-core/skills/config/eslint/templates/base.js"
      }
    }
  ],
  "summary": {
    "total_discrepancies": 1,
    "apply_count": 1,
    "update_template_count": 0,
    "ignore_count": 0,
    "custom_count": 0
  }
}

---

Edge Cases

Scenario	Behavior
Zero discrepancies	Skip this phase, proceed to report
User enters invalid choice	Re-prompt with error message
User enters [4] without instruction	Prompt for instruction text
Same file multiple discrepancies	Present each separately, allow different decisions
Critical severity	Highlight in presentation, but still user decides

---

Integration Points

Inputs from: /skill audit-investigation Outputs to: /skill audit-remediation, /skill template-update

Phase order:

1. Investigation (READ-ONLY, collects findings)
2. Resolution (HITL, captures decisions) ← This skill
3. Remediation (WRITE, applies decisions)