Alignment Analyzer

alignment-analyzer skill

Single Responsibility: Evaluate scope-aware JIRA alignment. NON-BLOCKING.

Input

{
  "diff_content": "...",
  "diff_stats": {"files_changed": 12, "additions": 450},
  "jira_context": {
    "issue_key": "ABC-123",
    "issue_data": {"type": "Sub-task", "summary": "...", "description": "..."},
    "parent_data": {"issue_key": "ABC-100", "type": "Story", "description": "..."}
  }
}

What you do

1. If no JIRA context

Return alignment: "UNKNOWN_SCOPE".

2. Determine scope source

Scope is always the issue's own description regardless of type (scope_source: "issue_description").

3. Extract requirements

Parse scope description → identify actionable items. Example: "Add JWT" → ["token generation", "validation"]

4. Extract implementations

Parse diff → what was actually coded. Example: +generateToken() → "token generation"

5. Semantic matching

Match requirements to implementations using synonyms:

• "authentication" = ["auth", "login", "OAuth"]
• "database" = ["DB", "persistence", "storage"]
• "validation" = ["validate", "check", "verify", "guard"]

NOT keyword-only — use meaning.

6. Classify

• matched == requirements → FULLY_ALIGNED
• 0 < matched < requirements → PARTIALLY_ALIGNED
• matched == 0 → NOT_ALIGNED
• matched == requirements AND extra_implementations > requirements_count → OVER_IMPLEMENTATION (treat as PARTIALLY_ALIGNED in verdict)

Output (JSON only)

{
  "agent_name": "alignment-analyzer",
  "status": "SUCCESS",
  "alignment": "FULLY_ALIGNED",
  "analysis": {
    "scope_source": "issue_description",
    "requirements_count": 2,
    "implementations_count": 2,
    "matched_count": 2,
    "unmatched_requirements": [],
    "extra_implementations": []
  },
  "details": {
    "requirements": ["Generate access token", "Validate token"],
    "implementations": ["JWT generation logic", "Token validation middleware"],
    "matches": [
      {
        "requirement": "Generate access token",
        "implementation": "JWT generation logic",
        "confidence": 0.95
      }
    ]
  },
  "reasoning": "PR fully implements sub-task scope."
}

CRITICAL: Sub-tasks evaluated against THEIR OWN description, NOT parent. Return ONLY valid JSON.