{
  "meta": {
    "agent": "planner",
    "task_id": "483",
    "title": "Fix PM skill auto-complete flakiness under auto directive",
    "created_at": "2026-04-13T12:00:00Z"
  },
  "inputs": [
    {
      "name": "issue-483",
      "type": "context",
      "ref": "https://github.com/AgentSDE/agent-core/issues/483",
      "notes": "PM skill intermittently blocks with PERSISTENT instead of auto-completing"
    },
    {
      "name": "SKILL.md",
      "type": "file",
      "ref": ".claude/skills/product-manager/SKILL.md",
      "notes": "Main skill prompt — Steps 6 and 8 need directive-aware logic"
    },
    {
      "name": "role.md",
      "type": "file",
      "ref": ".claude/skills/product-manager/role.md",
      "notes": "Role rules — needs auto-complete directive rules"
    }
  ],
  "outputs": [
    {
      "name": "plan",
      "type": "plan",
      "format": "md",
      "content": "plan.md"
    }
  ],
  "files": [
    {
      "path": ".claude/skills/product-manager/SKILL.md",
      "action": "modify",
      "reason": "Add directive-aware auto-completion logic to Steps 6 and 8"
    },
    {
      "path": ".claude/skills/product-manager/role.md",
      "action": "modify",
      "reason": "Add Do/Do Not rules for auto-directive behavior"
    }
  ],
  "steps": [
    {
      "id": "S1",
      "summary": "Add directive-aware auto-completion logic to SKILL.md (Steps 6 and 8) and update role.md rules",
      "acceptance": [
        "SKILL.md Step 6 classifies issues as well-specified when DIRECTIVE=auto and issue has clear problem statement + acceptance criteria",
        "SKILL.md Step 8 skips question-asking and proceeds to PRD finalization for well-specified issues under auto directive",
        "SKILL.md Step 8 includes fallback: genuinely ambiguous issues (empty body) still get 1 round of questions under auto, then force finalization",
        "role.md has Do rule for auto-completing well-specified issues under auto directive",
        "role.md has Do Not rule against asking questions for well-specified issues under auto directive",
        "Careful directive behavior is unchanged — existing question-asking logic preserved"
      ],
      "depends_on": []
    }
  ],
  "risks": [
    {
      "risk": "Prompt changes are non-deterministic — Claude may still occasionally ask questions",
      "mitigation": "Use strong directive language (MUST, NEVER) in the prompt"
    },
    {
      "risk": "Over-aggressive auto-completion for borderline issues",
      "mitigation": "Genuinely ambiguous fallback ensures empty/nonsensical issues still get questions"
    }
  ],
  "assumptions": [
    "The DIRECTIVE value is accessible within the skill execution context (passed via arguments or task.json)",
    "Two files is sufficient scope — no pipeline code changes needed per issue scope"
  ],
  "open_questions": []
}