OpenEngine

Verify my Linear connection before we continue setting up Open Engine.

Do these checks in order:
1. List the Linear workspaces, teams, or projects you can see.
2. Tell me which Linear account appears to be connected.
3. Find or create one throwaway test issue named [agent instructions][connection-test][task] Verify Linear MCP access.
4. Add a comment to that issue saying AGENT CONNECTION TEST from this runtime.
5. Move or update only that test issue if you have status-write access.
6. Report whether you have read access, comment access, and status-update access.

Do not touch any real work issues during this test.

---
name: open-agent-engine
description: Route assigned Linear agent tasks through the Open Agent Engine queue.
version: 1
---

Agent code: alex-codex
Human/operator: Alex Example
Linear team: Agent Engine
Linear project: Personal Agent Engine
Agent label: agent-instructions
Status ledger issue: ENG-000   # placeholder until you create the ledger in Step 4
Private sources: list approved local folders or private issue ids

Rules:
- Process at most one eligible task issue per run.
- Only process issues assigned to this human/operator.
- Only process issues with the agent-instructions label and [agent instructions] in the title.
- Only claim task issues whose second title bracket is this agent code, e.g. [agent instructions][alex-codex][task]. Standing issues addressed to [all agents] still apply to every runtime.
- Before task work, check standing context versions assigned to this agent or all agents.
- Claim by moving the issue to Agent Working and leaving AGENT CLAIMED.
- Re-read the issue after claiming.
- If complete with no human judgment required, leave AGENT DONE and move to Agent Done.
- If complete but review, QA, approval, or inspection is required, leave AGENT DONE and move to Agent Review.
- If the missing answer belongs on Linear, ask one specific question, leave AGENT BLOCKED, and move to Agent Needs Input.
- If the question is about local runtime permissions, skill install approval, automation setup, account authority, or private agent-thread context, ask in the human's own agent thread/app, leave AGENT HUMAN HOLD, and move to Agent Needs Input.
- Ask before publishing, emailing, posting publicly, deploying, changing billing, changing credentials, deleting destructive data, or making customer-facing changes.

<task>
  Create the private standing setup issue for our Open Engine.
</task>

<issue>
  Title: [agent instructions][all agents][standing_skill] Install <Engine Name> core context v<version>
  Label: agent-instructions
  Status: Standing
</issue>

<include>
  <item>What this engine is for.</item>
  <item>Status ledger issue ID or URL.</item>
  <item>Routing map issue ID or URL.</item>
  <item>Private context packs to install or adapt.</item>
  <item>Codex setup steps plus manual fallback.</item>
  <item>Other runtime setup notes where known.</item>
  <item>Required AGENT AUTOMATION READY receipt after install and smoke test.</item>
</include>

<privacy>
  Keep org charts, brand guides, customer context, secrets, account details, and private skill bodies inside this private issue or local runtime context.
</privacy>

AGENT STATUS
Agent: <agent-code>
Human/operator: <name or unknown>
Runtime: <Codex | Claude | Grok | other>
Automation: <automation name or manual>
Automation state: <installed | manual-required | blocked | paused>
Last heartbeat: <ISO8601 timestamp>
Last queue result: <checking | none | observed ISSUE-123 | claimed ISSUE-123 | completed ISSUE-123 | blocked ISSUE-123 | holding ISSUE-123 | resumed ISSUE-123 | failed ISSUE-123>
Last successful run: <ISO8601 timestamp or unknown>
Local context: <engine version>; <routing map version>
Notes: <none or short blocker>

<task>
  Run one Open Engine queue check for this operator.
</task>

<order>
  <step>Identify this runtime's agent code.</step>
  <step>Open the status ledger issue.</step>
  <step>Find this agent's top-level AGENT STATUS comment.</step>
  <step>Update that comment in place with Last queue result: checking and the current timestamp.</step>
  <step>Run standing context preflight before new task work.</step>
  <step>Check AGENT HUMAN HOLD issues before new task work. If a held issue now shows AGENT HUMAN ANSWERED, move it back to Agent Working, leave AGENT RESUMED, finish it, and stop after this one issue.</step>
  <step>Check AGENT BLOCKED issues before new task work. If a blocked issue now has its answer on the same issue, move it back to Agent Working, leave AGENT UNBLOCKED then AGENT RESUMED, finish it, and stop after this one issue.</step>
  <step>Check delegated issues this agent routed to someone else.</step>
  <step>If no hold or blocked issue is ready to resume, find the oldest eligible Agent Todo task assigned to this operator.</step>
  <step>Eligible tasks must have the agent-instructions label, [agent instructions] in the title, and this runtime's agent code as the second title bracket, e.g. [agent instructions][alex-codex][task].</step>
  <step>If no eligible issue exists, update the ledger with Last queue result: none and stop.</step>
  <step>If an issue exists, move it to Agent Working and leave AGENT CLAIMED.</step>
  <step>Re-read the issue after claiming.</step>
  <step>Do only the scoped work.</step>
  <step>If complete with no human judgment needed, leave AGENT DONE and move to Agent Done.</step>
  <step>If complete but review, QA, approval, inspection, or publishing is needed, leave AGENT DONE and move to Agent Review.</step>
  <step>If the missing answer belongs on Linear, ask one specific question, leave AGENT BLOCKED, move to Agent Needs Input, update the ledger with blocked ISSUE-ID, and stop.</step>
  <step>If the answer belongs in the human's own agent thread/app, ask there, leave AGENT HUMAN HOLD, move to Agent Needs Input, update the ledger with holding ISSUE-ID, and stop.</step>
  <step>If execution fails unexpectedly, leave AGENT FAILED with the last safe step and retry count.</step>
  <step>Update the status ledger with completed, blocked, holding, resumed, failed, observed, or claimed for the issue id.</step>
  <step>Stop after exactly one task issue.</step>
</order>

<receipts>
  Use AGENT CLAIMED before work, AGENT DONE when complete, AGENT BLOCKED for Linear-answerable blockers, AGENT HUMAN HOLD for owner-thread permissions, AGENT UNBLOCKED then AGENT RESUMED when a blocked issue's answer arrives, AGENT HUMAN ANSWERED then AGENT RESUMED when a hold is answered, and AGENT FAILED only for unrecoverable failure.
</receipts>

<boundaries>
  Never publish, email, Slack-post, deploy, delete, change billing, change credentials, or make outward-facing changes unless the issue explicitly grants that approval.
</boundaries>

## Requester
<your name>.

## Desired outcome
Leave a short comment proving the queue runner claimed and completed this issue.

## Sources
None.

## Acceptance criteria
- Issue has AGENT CLAIMED from <your-agent-code>.
- Issue has AGENT DONE from <your-agent-code>.
- Issue moved to Agent Done.
- Status ledger says completed for this issue id.

## Boundaries
Do not edit files, publish, email, Slack-post, deploy, delete, or change credentials.

## If blocked
If the missing answer belongs on Linear, leave AGENT BLOCKED with one specific question.
If the missing answer belongs in the human's own agent thread/app, leave AGENT HUMAN HOLD and wait there.

## Agent routing map

Alex Example
- Linear assignee: Alex Example
- Agent codes: alex-codex, alex-claude
- Route to Alex for: engineering, local repo work, QA

Sam Example
- Linear assignee: Sam Example
- Agent codes: sam-codex
- Route to Sam for: editorial review, social drafts

Rules:
- If assigning work to Sam's agent, assign the issue to Sam.
- If the target agent is not online in the status ledger, say that before relying on the handoff.
- If an agent is not listed yet, it may propose a unique agent code, ask its human to fill in the routing details, and leave those details as a comment on this routing map issue.
- Human approval is required for publishing and customer-facing changes.