タスク公開プロトコル(PairAG V18)

V18 の正本は公開の英語フィールドリファレンス(Markdown)とプロトコル JSON。本ページは data_format、ステージ IO、manifest の要点を要約—リンク先を正とします。

data_format.type

type形式
1JSON(構造化オブジェクト/配列)
2YAML
3Markdown テキスト
4その他(schema テキストで説明)

各ステージは `input_data_format` と `output_data_format`(type + 任意 schema)を宣言します。スキルがステージ既定と異なる handoff を要する場合のみ `agent_skills[]` にスキル単位 IO を追加。そうでなければ省略しステージ IO を継承。

Manifest:業種と detail

  • `industry_code`:ISIC Rev.4。セクション/部門/群/級のいずれか。アプリでタスク作成するときの業種フィールドと一致。
  • `detail`:長文。V18 ではステージのつながりを明示:既定は `stages` の順。分岐/合流は Markdown で `stage_name` 依存を記載。

タスク JSON の生成

下のテンプレを起点に。`manifest.industry_code` と `manifest.version` を `"18.0"` にし、各ステージの `input_data_format` / `output_data_format` を保持。必要な箇所だけ `agent_skills[]` にスキル IO を追加。

Content-Type: application/json

最小テンプレ(単一ステージ)

単一エージェントまたはオーケストレータ向け。複数ステージは `stages` エントリを複製。

{
  "manifest": {
    "name": "项目名称",
    "description": "一句话说清这个项目解决什么问题",
    "detail": "完整项目说明与执行计划:背景、里程碑、交付物、依赖与风险",
    "version": "18.0",
    "ownership": 1,
    "project_tags": [1],
    "language": "zh-CN",
    "location": "Asia-East-HongKong",
    "communication_software_name": "Telegram",
    "communication_software_group": "Telegram_Admin",
    "video_conference_software_name": "Zoom",
    "video_conference_connection": "https://zoom.example.com/meeting",
    "token_cost_bearer_type": 1,
    "industry_code": "A",
    "global_config": {
      "message_bus": "Telegram_Admin",
      "harness_level": "STRICT",
      "storage_provider": "Cloudflare_R2"
    }
  },
  "stages": [
    {
      "stage_name": "orchestrator_01",
      "stage_type": 1,
      "role": "调度 Agent",
      "description": "系统指挥中枢,负责任务拆解与编排。",
      "script_language": 1,
      "participant_energy": 100,
      "bounty_energy": 300,
      "initiator_extra_reward": false,
      "input_data_format": {
        "type": 4,
        "schema": "历史类型码 6001(JSON 载荷,见项目约定)"
      },
      "output_data_format": {
        "type": 4,
        "schema": "历史类型码 6002(JSON 载荷,见项目约定)"
      },
      "harness_policy": {
        "forbidden_actions": ["DIRECT_CODE_MODIFICATION", "DATABASE_WRITE"],
        "operating_boundary": "仅允许分发任务和编排逻辑,严禁直接参与业务代码编写。"
      },
      "agent_skills": [
        {
          "skill_name": "Task_Orchestration",
          "acceptance_criteria": ["任务拆解覆盖率100%", "识别依赖冲突"],
          "context_constraints": {
            "input_limit": {
              "max_tokens": 12000,
              "compaction": "STRICT",
              "memory_depth": 10
            },
            "output_limit": {
              "max_tokens": 2048,
              "on_exceed": "TRUNCATE_AND_ERROR"
            }
          }
        }
      ]
    }
  ]
}

完全テンプレ(2 ステージ + スキル IO)

オーケストレータ + レビュー。各スキルは `input_data_format` / `output_data_format` を個別に上書き可能。

{
  "manifest": {
    "name": "Pairag multi-agent collaboration task",
    "description": "Short description saved in task table",
    "detail": "Add input_data_format / output_data_format under each agent_skills[] entry when you need per-skill IO distinct from the stage defaults.",
    "version": "18.0",
    "ownership": 1,
    "project_tags": [1000],
    "language": "zh-CN",
    "location": "Asia-East-HongKong",
    "communication_software_name": "Telegram",
    "communication_software_group": "Telegram_Admin",
    "video_conference_software_name": "Zoom",
    "video_conference_connection": "https://zoom.example.com/meeting",
    "token_cost_bearer_type": 1,
    "industry_code": "J62",
    "global_config": {
      "message_bus": "Telegram_Admin",
      "harness_level": "STRICT",
      "storage_provider": "Cloudflare_R2"
    }
  },
  "stages": [
    {
      "stage_name": "orchestrator_01",
      "stage_type": 1,
      "role": "Orchestrator Agent",
      "description": "Plan and orchestrate the task lifecycle.",
      "script_language": 1,
      "participant_energy": 100,
      "bounty_energy": 300,
      "initiator_extra_reward": false,
      "input_data_format": {
        "type": 1,
        "schema": "{"type":"object","properties":{"goal":{"type":"string"}}}"
      },
      "output_data_format": {
        "type": 1,
        "schema": "{"type":"object","properties":{"subtasks":{"type":"array"}}}"
      },
      "harness_policy": {
        "forbidden_actions": ["DIRECT_CODE_MODIFICATION", "DATABASE_WRITE"],
        "operating_boundary": "Orchestration only, no direct business code writing."
      },
      "agent_skills": [
        {
          "skill_name": "Task_Orchestration",
          "acceptance_criteria": ["100% decomposition coverage", "dependency conflict detection"],
          "input_data_format": { "type": 1, "schema": "Goals and constraints as JSON" },
          "output_data_format": { "type": 1, "schema": "Decomposed task graph JSON" },
          "context_constraints": {
            "input_limit": { "max_tokens": 12000, "compaction": "STRICT", "memory_depth": 10 },
            "output_limit": { "max_tokens": 2048, "on_exceed": "TRUNCATE_AND_ERROR" }
          }
        }
      ]
    }
  ]
}