本文へ移動
cccskills
無料GitHub で公開日本語紹介

opc-collab

AutoSwarmの会社モードで、エージェントへの連絡や作業の委任、会議への応答、管理ボードの確認、タスク調整の提案を行います。

原文原文の説明を見る

Collaborate with other agents in AutoSwarm company mode via the `opc-collab` CLI — send messages, request user input, delegate work, read the manager board, respond in meetings, propose task adjustments.

HKUDS/AutoSwarmスター 1,963(GitHub 1,963 + cccskills 0)
連携ツールPostgreSQL

実行スコア 未検証

インストール方法を見る

こんなときに便利

  • チームメンバーに作業状況を連絡したいとき
  • 不足情報をユーザーに確認したいとき
  • 担当者へ作業を委任したいとき
  • 管理ボードで子タスクを確認したいとき
  • 会議で意見や決定を返したいとき

日本語での紹介

できること

AutoSwarmの会社モードで、同じ大きな作業に参加するエージェントと連携します。opc-collab CLIを使って、個別メッセージや複数の役割への課題共有、返信、ユーザーへの入力依頼を行えます。管理者の立場では、子タスクの委任や管理ボードの確認、既存タスクの修正・削除、レビューへの対応ができます。会議を開始して結果を待つ、会議に返答する、実行中のタスク計画の変更を提案するといった操作も対象です。

こんなときに便利

複数のエージェントで一つの作業を進め、担当者間の連絡や引き継ぎが必要なときに向いています。作業がユーザーの判断待ちになった場合や、管理者として仕事を分割し、進捗や依存関係を確認したい場合にも使えます。会議で決定をまとめる場面にも役立ちます。

使い方の例

  • 「レビュー担当に草稿の確認を依頼してください」
  • 「この作業を準備と最終化に分けて担当者へ委任してください」
  • 「管理ボードを確認して、会議の議題に返信してください」

注意点

AutoSwarmの会社モードで起動された実行環境が必要です。opc-collab CLIをシェルから呼び出し、実行時に設定された環境変数から本人・プロジェクト・タスク情報を取得します。利用できる操作は役割や実行段階によって異なります。

この紹介文は、公開されている SKILL.md をもとに AI(OpenAI Codex)が作成しました。正確な仕様は下の原文を確認してください。

含まれるファイル(1)

  • SKILL.md14.7 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

AutoSwarm company-mode collaboration

You are running inside an AutoSwarm "company mode" run as one of several agents collaborating on a larger task. This skill gives you a local command, opc-collab, that you invoke through the shell to talk to teammates, delegate child work, and read your manager board.

The CLI is on your PATH. Every invocation reads its identity (who you are, which project/session, which task) from environment variables that AutoSwarm already set for you before launching this process — you do NOT pass those.

When to use this skill

  • You need to message another role (DM, broadcast, or blocking question).
  • Your own current work item is blocked on a user decision or missing input.
  • You are a manager and need to delegate child work items, or inspect your kanban board.
  • You are in a meeting and need to respond or finalize a decision.
  • You want to propose a runtime task adjustment (replan).

Do not use this CLI for general shell work, file edits, or code changes — those go through your native tools (Bash, Edit, Read, etc.).

How to call it

General form:

opc-collab <tool> --args-json-file args.json

All arguments go in the JSON object. The CLI prints a JSON result on stdout and exits 0 on success, 1 on error (with the error message on stderr and a {"error": "..."} body on stdout).

Prefer --args-json-file or --args-stdin over inline JSON. They are unambiguous for nested objects and arrays, and they work consistently on Linux, macOS, Windows PowerShell, and Windows CMD. If OPC_COLLAB_CLI is set, use that executable path; otherwise use opc-collab from PATH. In AutoSwarm-spawned Windows runs, do not use --args-json or pipe JSON into --args-stdin; command-line and PowerShell pipeline text can corrupt non-ASCII before it reaches the CLI. Write the JSON object to a UTF-8 file and call opc-collab <tool> --args-json-file <file> instead. PowerShell-safe UTF-8 file write: $enc = New-Object System.Text.UTF8Encoding $false; [System.IO.File]::WriteAllText($path, $json, $enc).

Available tools

Not every tool is available every turn. The runtime narrows the surface based on your role and phase; if you call a tool that is not allowed, the CLI prints an error telling you which tools ARE allowed.

Messaging

  • inbox — check or acknowledge your mailbox without implicitly marking messages read. Use status for counts, peek for actionable message bodies, and ack only after handling messages that do not need a reply. reply_message automatically acknowledges the original message. If a manager-board action already handled the matching approval/review request, acknowledge that inbox message with ack.

    {
      "action": "status"
    }
    
    opc-collab inbox --args-json-file args.json
    
    {
      "action": "ack",
      "message_ids": ["<msg_id>"]
    }
    
    opc-collab inbox --args-json-file args.json
    
  • send_dm — async direct message.

    {
      "to_agent": "reviewer",
      "subject": "Draft ready for review",
      "body": "The v1 draft is in ./deliverables/draft.md. Please review."
    }
    
    opc-collab send_dm --args-json-file args.json
    
  • ask_peer_and_wait — blocking peer question; pauses this run until the peer replies or the timeout fires.

    {
      "to_agent": "cto",
      "subject": "Which DB shall I target?",
      "body": "Postgres or SQLite for the prototype?",
      "timeout_seconds": 300,
      "on_timeout": "continue"
    }
    
    opc-collab ask_peer_and_wait --args-json-file args.json
    
  • reply_message — reply to a received message. If the reply is rejected by policy or transport but the message has already been handled through the board or current task result, use inbox with action="ack" for that message.

    {
      "message_id": "<msg_id from your inbox>",
      "body": "Go with Postgres — we already have it in prod."
    }
    
    opc-collab reply_message --args-json-file args.json
    
  • broadcast_issue — raise an async issue to multiple roles.

    {
      "to_agents": ["cto", "cmo"],
      "subject": "Blocked on pricing policy",
      "body": "Need a decision before I can land the checkout flow."
    }
    
    opc-collab broadcast_issue --args-json-file args.json
    
  • request_user_input — pause your own current work item and ask the user for missing input. This tool is self-scoped: do not pass target_role, target_task_id, work_item_id, or any other target id. AutoSwarm reads your role, task, seat, and work item from the runtime environment. After it returns requires_user_input=true, stop the current turn and wait for the user's reply.

    {
      "reason": "Blocked on a deployment choice that only the user can make.",
      "questions": [
        {
          "id": "deployment_region",
          "header": "Deployment region",
          "question": "Which deployment region should I target?",
          "options": [
            {"label": "US East", "description": "Use us-east-1"},
            {"label": "EU West", "description": "Use eu-west-1"},
            {"label": "Asia", "description": "Use ap-east-1"}
          ],
          "allow_freeform": true,
          "required": true
        }
      ],
      "context_note": "I can continue once the region is selected."
    }
    
    opc-collab request_user_input --args-json-file args.json
    

    Each question can have up to three selectable options. Option ids default to a, b, and c; the UI also offers Other/freeform unless allow_freeform=false. Legacy "questions": ["..."] remains valid and produces a freeform-only review card.

Manager / delegation

Leader Delegation Planning Overlay

Before delegate_work, managers should convert the upstream assignment into self-contained child work packets. This overlay does not replace your role prompt; it only governs delegation quality.

Use this checklist before dispatching:

  • Preserve upstream intent.

  • Determine local leadership scope.

  • Map deliverables to owners.

  • Decompose into work items.

  • Distinguish hard dependencies from can-start-now prework.

  • Split one role into multiple phase items when that unlocks progress; do not split mechanically.

  • Add dependencies / non-overlap boundaries.

  • Fill planning_context.

  • Fill per-item brief, outputs / deliverables, done_when / acceptance_criteria, delegation_rationale, and non_overlap_guard.

  • delegate_work — as a manager, create child work items for your direct reports. Every item needs role_id, title, and a complete brief (legacy summary is accepted). scope_key is optional; the runtime can generate one. depends_on can name a sibling role, scope key, work item ref, or work item id. The same role_id may appear in multiple items when each item has a distinct deliverable, phase, dependency shape, or handoff.

    {
      "planning_context": "Leader planning summary: upstream goal, local scope, deliverable map, can-start-now work, hard dependencies, sequencing, assumptions.",
      "items": [
        {
          "role_id": "target_role",
          "title": "Startable preparation slice",
          "brief": "Context: This phase can start before upstream dependencies finish. Mission: Produce preparation artifacts that unblock later finalization. Required outputs: Leave the concrete artifact or handoff named in outputs. Location / handoff: Use the agreed output path or report channel. Quality expectations: Meet the done_when criteria below. Boundaries: Stay inside this phase. Insufficient work: Do not submit only vague notes. Completion report: List artifacts, verification, assumptions, and blockers.",
          "scope_key": "target-role-prep",
          "work_kind": "execute",
          "outputs": ["Concrete preparation artifact or handoff"],
          "done_when": ["Preparation output is useful for dependent finalization"],
          "delegation_rationale": "Why this direct report owns this slice.",
          "non_overlap_guard": "This item owns preparation; a sibling item owns finalization.",
          "coordination_notes": "Record assumptions that finalization must revisit.",
          "depends_on": []
        },
        {
          "role_id": "target_role",
          "title": "Dependency-bound finalization slice",
          "brief": "Context: This sibling item finalizes the same role's output after dependencies land. Mission: Integrate dependency evidence into the final production deliverable. Required outputs: Leave the final artifact named in outputs. Location / handoff: Use the agreed output path or report channel. Quality expectations: Verify against prep and dependency inputs. Boundaries: Do not redo prep unless quality requires it. Insufficient work: Do not finish before dependencies are incorporated. Completion report: List final artifacts, verification, assumptions, and blockers.",
          "scope_key": "target-role-finalize",
          "work_kind": "execute",
          "outputs": ["Final dependency-aware artifact"],
          "done_when": ["Final output incorporates dependencies and is ready to integrate"],
          "delegation_rationale": "Same role owns final quality for this slice.",
          "non_overlap_guard": "This item owns finalization; sibling prep owns early scaffolding.",
          "coordination_notes": "Use precise scope_key dependencies when a role owns multiple items.",
          "depends_on": ["target-role-prep"]
        }
      ]
    }
    
    opc-collab delegate_work --args-json-file args.json
    

    brief is the child work item's full assignment packet, not a short summary. Do not use description. Recommended brief shape: Context, Mission, Required outputs, Location / handoff, Quality expectations, Boundaries, Insufficient work, Completion report. When the same role gets multiple sibling items, set stable scope_key values and reference those keys in depends_on; broad role references may become ambiguous.

  • manager_board_read — READ-ONLY view of your direct reports' child items. For your current manager board, omit parent_work_item_id; the runtime uses $OPC_WORK_ITEM_ID automatically. Pass an explicit id only when it is a FULL WorkItem id returned by a prior tool call (no truncation). Never pass $OPC_TASK_ID or $OPC_RUNTIME_TASK_ID; those are runtime Task ids, not WorkItem ids. After you approve, reject, or otherwise handle a child through the board review flow, acknowledge any matching approval/review inbox message unless reply_message already did it.

    {
      "include_children": true
    }
    
    opc-collab manager_board_read --args-json-file args.json
    

    Do NOT use legacy aliases like parent_id, include_outputs, scope, reason, note — the tool rejects them.

  • modify_work_item — revise an existing child WorkItem on your current manager board when the owner is still correct but the title, brief, deliverables, acceptance criteria, dependencies, or coordination fields are wrong. Read the board first, then pass the exact full work_item_id. Prefer this over creating a duplicate replacement item.

    {
      "work_item_id": "work-item-id-from-manager_board_read",
      "task_brief": "Revised concrete assignment for the same owner.",
      "deliverables": ["Updated artifact or handoff"],
      "acceptance_criteria": ["Updated acceptance condition"],
      "depends_on": ["sibling-scope-key-or-full-work-item-id"],
      "reason": "Why the existing card needed revision.",
      "reset_to_ready": true
    }
    
    opc-collab modify_work_item --args-json-file args.json
    
  • delete_work_item — cancel or hide an obsolete/wrong child WorkItem on your current manager board so it no longer blocks parent synthesis. Read the board first, then pass the exact full work_item_id.

    {
      "work_item_id": "work-item-id-from-manager_board_read",
      "reason": "Why this child card is obsolete or wrong.",
      "replacement_dependency_work_item_ids": []
    }
    
    opc-collab delete_work_item --args-json-file args.json
    
  • close_human_review — close the owner-facing delivery review when you decide the user's latest directive means the delivery is accepted and no further internal work is needed. Do not call this for requested changes; use board tools or reply directly instead.

    {
      "summary": "The user accepted the delivery; no further work is required.",
      "user_message": "Acknowledged. I am closing the human review for this delivery."
    }
    
    opc-collab close_human_review --args-json-file args.json
    

Meetings

  • start_meeting — open a meeting and wait for the outcome.
    {
      "topic": "Prioritization for Q2",
      "participants": ["cto", "cmo", "cfo"],
      "agenda": ["Goals", "Trade-offs", "Decision"]
    }
    
    opc-collab start_meeting --args-json-file args.json
    
  • respond_meeting — respond to a live meeting, optionally finalize it if you are the decision owner.

Other

  • propose_task_adjustment — propose a runtime replan (summary + changeset).
  • route_work — coordinator tool; send_followup / spawn_task / escalate.
  • read_inbox / read_meeting / list_colleagues — debug/admin reads that may not be available outside of debug mode.

Argument contract rules

  1. Use the exact argument names shown above. brief/summary, outputs/deliverables, and done_when/acceptance_criteria are the supported aliases; do not invent others.
  2. Work-item IDs are verbatim. When a prior response hands you a work_item_id, copy the full string unmodified into the next call. Never truncate.
  3. delegate_work items use brief or legacy summary, not description.
  4. manager_board_read takes only parent_work_item_id and optional include_children. No other args are accepted. For your current manager board, omit parent_work_item_id; never use $OPC_TASK_ID.

Error handling

  • Exit code 0 + JSON on stdout: success, use the payload.
  • Exit code 1 + {"error": "..."} on stdout, message on stderr: the call failed. Read the error and decide whether to retry with corrected args, escalate, or continue with what you have.
  • An error containing "not available for this run" means the runtime did not grant this tool for your current role/phase. The error message lists the tools you CAN call — pick one of those instead.

Only transport

opc-collab is the supported way to reach your teammates. If the CLI returns an error, inspect the error message and retry or adapt.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

コメント

ログインなしで書き込めます。使ってみた感想、質問、動かなかった環境などを気軽にどうぞ。

まだコメントはありません。最初のコメントを書いてみませんか。

送信するとすぐに公開されます。メールアドレスや電話番号などの個人情報は書かないでください。URL は 2 つまでです。

0 / 1000 文字

このフォームは reCAPTCHA で保護されており、Google のプライバシーポリシーと利用規約が適用されます。

同じリポジトリのスキル

clawhub

無料概要あり

ClawHubで目的に合うエージェントスキルを検索し、プロジェクトへの導入やインストール済みスキルの更新を進めるスキルです。

  • 作業に合うスキルを探したいとき
  • プロジェクトにスキルを導入したいとき
  • 導入済みスキルを更新・確認したいとき
HKUDS/AutoSwarm1,9632026年10月25日 更新

collaboration-playbook

無料概要あり

AutoSwarmの会社モードで、作業範囲の守り方や引き継ぎ、同僚への連絡、会議、共同作業が止まる場合の進め方を定めます。

  • 担当作業の引き継ぎを整えたいとき
  • 同僚へ連絡する条件を判断したいとき
  • 役割間の会議やブロック対応を整理したいとき
HKUDS/AutoSwarm1,9632026年10月25日 更新

cron

無料概要あり

システムのcrontabを使い、定期実行するタスクや一度だけ実行する予定を設定・確認・削除する方法を案内します。リマインダーや定型処理のスケジュール管理に使えます。

  • 定期リマインダーの設定
  • 一度だけの予定の登録
  • cronジョブの確認と削除
HKUDS/AutoSwarm1,9632026年10月25日 更新

github

無料概要あり

GitHubのIssueやプルリクエスト、CI実行状況を`gh`コマンドで確認し、必要に応じてAPIから詳細情報を取得するスキルです。

  • プルリクエストのCI状況を確認したいとき
  • 失敗したワークフローのログを調べたいとき
  • 最近のワークフロー実行を一覧したいとき
HKUDS/AutoSwarm1,9632026年10月25日 更新

memory

無料概要あり

ユーザーが後のセッションでも使いたいと確認した設定やプロジェクト固有の情報を、全体用・プロジェクト用のMarkdownメモに整理して保存します。

  • 継続して使う回答の好みを残したいとき
  • プロジェクトの規約や制約を記録したいとき
  • 次回以降に必要な作業パスを整理したいとき
HKUDS/AutoSwarm1,9632026年10月25日 更新

skill-creator

無料概要あり

エージェント向けスキルの目的や利用場面を整理し、SKILL.mdの構成、付属資料、命名、検証、配布パッケージ化まで作成・更新を案内します。

  • 繰り返し作業をスキルにまとめたいとき
  • SKILL.mdと付属資料の構成を決めたいとき
  • スキルを検証して配布用にまとめたいとき
HKUDS/AutoSwarm1,9632026年10月25日 更新

HKUDS のスキルをすべて見る

このスキルの問題を報告する