From flowstate-skills
Use when an agent needs another agent's help and must choose whether to assign a task, mention an agent, reply to an inbox item, open an agent conversation, or request approval - provides the inbox-aware routing table for agent-to-agent collaboration
How this skill is triggered — by the user, by Claude, or both
Slash command
/flowstate-skills:flowstate-agent-collaboration-routingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
**Status:** Active
Status: Active Purpose: Pick the smallest FlowState collaboration channel that will reliably surface in the target agent's inbox Scope: Agent-to-agent work coordination across tasks, discussions, conversations, approvals, and description mentions Trigger: An agent needs another agent to act, review, decide, answer, or take ownership Input: Desired outcome, target agent slug or role, related FlowState entity ID Output: One routed collaboration action with a traceable inbox path
The inbox is the receiving agent's source of truth. Route collaboration to the inbox section that matches the work shape:
Need work done -> create/assign task -> Active tasks
Need entity context -> post @mention discussion -> Mentions
Need answer thread -> reply to discussion -> Replies
Need live exchange -> start/message conversation -> Conversations
Need formal gate -> create approval -> Approvals
Need durable owner -> description @mention -> Description mentions
Use the most concrete channel available. If you need execution, assign a task. If you only need attention on an existing entity, mention the agent on that entity. If you need back-and-forth with turn control, use an agent conversation.
| Desired result | Create this | Receiver sees it under | Companion skill |
|---|---|---|---|
| Agent owns implementation or investigation work | MCP task-create + task-assign, or task-claim for self-ownership | Active tasks | flowstate-agent-task-assignment |
| Agent should inspect or comment on a task/project/milestone | MCP agent-mention or discussion-create containing @<agent-slug> | Mentions | flowstate-agent-mentioning |
| Agent should continue an existing discussion thread | MCP discussion-reply with parentId set to the thread root | Replies | flowstate-agent-handoff-response |
| Agent should answer a short direct question | Agent-chat conversation message | Conversations | flowstate-agent-conversation-protocol |
| Agent must approve, reject, or request revisions | Approval record with approverId set to target teamMemberId | Approvals | flowstate-approval-workflow |
| Agent should remain visibly associated with an entity description | Entity description containing @<agent-slug> | Description mentions | flowstate-agent-mentioning |
Do not create multiple channels for the same ask unless each channel has a different job. A task assignment may include one discussion comment for context, but the task remains the action item.
Before writing anything that names another agent:
flowstate-agent-collaborator-lookup or MCP agent-resolve.teammembers.id.isAgent === true.@mentions.Never infer a team member ID from a display name, role title, or local agent file.
| Situation | Route |
|---|---|
| The target agent must spend real time producing an outcome | Create or update a task, assign owner, and mention them in a discussion |
| The target agent only needs to read and respond | Post an @<slug> discussion on the related entity |
| The target agent is already in a thread | Reply to the existing root discussion rather than starting a new thread |
| The target agent must unblock a process step | Create or update an approval, then mention the approval's related entity if needed |
| Two or more agents need bounded deliberation | Start an agent-chat conversation with maxTurns and expectedConfidence |
| A manager needs visibility into a subordinate's load | Use subordinate report/inbox commands before assigning more work |
After creating the routed item, run the lowest-cost verification:
| Channel | Verification |
|---|---|
| Task assignment | flowstate mcp call agent-inbox --json '{"slug":"<target-slug>","sinceDays":1,"limit":25}' --raw |
| Mention | Check the agent-mention/discussion-create response, or fallback collection-query discussions, and confirm mentions includes target |
| Reply | Check the discussion-reply response, or fallback collection-get discussions <reply_id>, and confirm parentId is the root discussion |
| Conversation | flowstate agent conversation show <conv_id> |
| Approval | collection-get approvals <approval_id> and confirm status: "pending" plus target approverId |
| Description mention | `collection-get <tasks |
If verification fails, fix the FlowState record. Do not tell the requester the agent was notified until the item is visible through the expected channel.
| Situation | Action |
|---|---|
| Target slug is unknown | Stop and list valid agents with flowstate agent list; do not create anonymous work |
| Target is a human team member | Use human assignment or approval flow; do not label it agent-to-agent collaboration |
| Related entity is missing | Create or recover the parent entity first; do not post orphaned context |
| Same ask already exists in receiver inbox | Update or reply to the existing item instead of duplicating |
Receiver has an In Progress task | Assign only urgent work, and state priority/blocker context in the task or mention |
| Item | Convention |
|---|---|
| Agent handle | @<slug> |
| Agent identity source | teammembers row resolved by slug |
| Work owner | tasks.assigneeId plus entityassignments owner |
| Runner owner | metadata.assignedAgent set to target teamMemberId |
| Traceability | Every collaboration write references a task, project, milestone, approval, or conversation ID |
| Inbox proof | Verify with target inbox, discussion query, conversation show, or approval get before claiming routed |
flowstate-agent-inbox - receiver-side triage of every routed itemflowstate-agent-collaborator-lookup - resolving agent slugs and team member IDsflowstate-agent-task-assignment - creating work that appears in Active tasksflowstate-agent-mentioning - writing mentions that index correctlyflowstate-agent-handoff-response - replying and escalating from inbox itemsflowstate-agent-conversation-protocol - bounded direct agent conversationsCreated: 2026-05-26
npx claudepluginhub epic-digital-im/epic-flowstate-skills --plugin flowstate-skillsGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.