Opc Collab logo

Opc Collab

OrganizationPopular
HKUDS
opc-collab

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

Overview

PublisherHKUDS
RepositoryOpenOPC
Skill nameopc-collab
Stars
1.7K
Forks
316
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by HKUDS on GitHub. Read the source before you install it.

Installation

Install the Opc Collab AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/HKUDS/OpenOPC.git /tmp/OpenOPC
mkdir -p .claude/skills
cp -r /tmp/OpenOPC/opc/skills_assets/opc_collab .claude/skills/opc-collab
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Opc Collab in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Opc Collab on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Opc Collab is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

OpenOPC company-mode collaboration

You are running inside an OpenOPC "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 OpenOPC 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:

bash
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 OpenOPC-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.

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

    json
    {
      "to_agent": "reviewer",
      "subject": "Draft ready for review",
      "body": "The v1 draft is in ./deliverables/draft.md. Please review."
    }
    bash
    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.

    json
    {
      "to_agent": "cto",
      "subject": "Which DB shall I target?",
      "body": "Postgres or SQLite for the prototype?",
      "timeout_seconds": 300,
      "on_timeout": "continue"
    }
    bash
    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.

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

    json
    {
      "to_agents": ["cto", "cmo"],
      "subject": "Blocked on pricing policy",
      "body": "Need a decision before I can land the checkout flow."
    }
    bash
    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. OpenOPC 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.

    json
    {
      "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."
    }
    bash
    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.

    json
    {
      "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"]
        }
      ]
    }
    bash
    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.

    json
    {
      "include_children": true
    }
    bash
    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.

    json
    {
      "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
    }
    bash
    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.

    json
    {
      "work_item_id": "work-item-id-from-manager_board_read",
      "reason": "Why this child card is obsolete or wrong.",
      "replacement_dependency_work_item_ids": []
    }
    bash
    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.

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

Meetings

  • start_meeting — open a meeting and wait for the outcome.
    json
    {
      "topic": "Prioritization for Q2",
      "participants": ["cto", "cmo", "cfo"],
      "agenda": ["Goals", "Trade-offs", "Decision"]
    }
    bash
    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.

Frequently asked questions

What does the Opc Collab AI skill do?

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

Why use Opc Collab on TypingMind?

Because you install it once and use it with any model. Opc Collab is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Opc Collab in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/HKUDS/OpenOPC/tree/main/opc/skills_assets/opc_collab. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Opc Collab?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Opc Collab?

As many as you like. As long as a model supports skills, you can use Opc Collab with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Opc Collab AI skill free?

Yes. It is published on GitHub by HKUDS under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇