Updates during a conversation

CLIO can receive an update while it is working: a background check finishes, a scheduled instruction becomes due, or an earlier decision is brought back into context. These updates appear in the activity record as Injection rows. Open one when you want to understand why the agent continued or changed direction.

Activity collection labelled Received loop wake, ran commands, containing an Injection Loop wake row followed by Run — An injection belongs to the activity collection. Here, a loop wake supplies the instruction and the Run row records the command used to check the files.

Understand an update

Expand the activity collection and find the syringe icon followed by Injection and its description. Open the row to read the supplied text. Use Open exact details for the recorded data when you need to inspect the full record.

The same activity collection with Injection Loop wake expanded to show its instruction and Open exact details — Open Injection Loop wake to read what the agent received. Open exact details exposes the recorded data; expanding the row does not run the command again.

An injection is not a message you typed or the agent’s final answer. Its presence also does not mean the requested work succeeded. A background completion notice delivers an outcome for the agent to inspect; the task’s actual result provides the evidence.

Common updates and what they mean

Type	Why CLIO supplies it
Background task completion; Results from background tasks	A previously started operation ended, or its retained result is being delivered.
Todo list reminder	Keeps the current checklist available to the agent.
Recovered conversation context; Summarization	Restores earlier context or carries a condensed record forward.
Memory search results	Provides relevant retained information found through memory search.
Plan mode; Replanning suggestion	Supplies the rules of Plan mode, or suggests reconsidering the plan after a stall.
Session plan; Session plan (shortened)	Offers the session plan while it is a draft, approved or active: when it changed and after a compaction. A plan too long for its share of the context is cut, with where to read the rest.
Plan edited by you	Offers the plan as you edited it in Work, once, at the agent’s next step, as your direction.
Plan not updated lately	One line when an active plan has gone 25 steps without an update. The agent decides whether it needs one.
Explore suggestion	After several reading calls in one response, notes that a separate explore agent could take the rest of the search. At most once per response.
Session facts	The conversation’s own facts: its workspace folder, where to write, the computer, its scratch folder, connected sources and uploaded files. Offered when they change and after a compaction.
Report instructions	Tells a subagent that its final answer goes to the agent that started it, as written.
Compaction failed; Compaction skipped	A notice for you that summarizing the context failed, or that the next automatic compaction waits for more room. The agent does not see these.
Notice from a background task; Message from a script	A message sent by a running task, such as a background script, naming the task. Show the script in Work opens its row.
Task result; Answer to a question	Heads each result or answer the agent receives with who sent it, the task and its status.
Goal judgment	Reports the goal check and guidance for continued work.
Loop wake; Scheduled wake	Delivers an instruction from a monitoring loop or scheduled action.
Repeated-failure warning	Calls attention to repeated unsuccessful operations.
Large result saved to a file	A tool result longer than the agent is shown: the agent reads its beginning, and the note says how long it was and which file holds the rest.
Path suggestion	Supplies a relevant file or folder location.
Hook	Supplies context produced by a configured hook.
Tool use	Supplies guidance about using the available tools.
Drafting alternatives; Advice for this draft	Supplies instructions or feedback while generating and reviewing alternatives.

Not every conversation uses every type. Goals, loops, schedules, hooks, and connected tools determine which additions are relevant.

Reminders

Most of these updates are reminders: short texts CLIO adds to what the agent reads, each with a rule for when it appears. Open Settings → Prompts and commands and choose the Reminders tab to see them. Each row has its title, what it does, when it appears, any settings such as the number of steps before the stale-plan note, and whether it is On or Off.

Switch one off with its switch. It is no longer added from then on. A switch keeps following the built-in wording, so later improvements to the text still reach you.
Edit its text by opening it. Your change is saved as a visible override of the built-in text, and the editor’s Rendered preview shows the result against the connected service.
Plugins’ reminders are listed beside CLIO’s own and can be switched off the same way. A plugin cannot make its reminder always on.
Some are always on. A reminder the agent cannot work correctly without, such as the note that a tool result was cut or the heading of each notification, shows Always on and a line saying why it cannot be turned off.

Example: a goal still needs work

A Goal judgment can explain why work should continue. Read the reason before treating a finished turn as a finished assignment. Here, two files had confirmed saves and the third still needed to be saved.

Expanded Goal judgment explaining that plan.md and results.md were saved but report.md still needed a successful save — The judgment names the missing deliverable and carries the goal's continuation instructions. It records the controller's guidance at that point in the workflow.

Example: a task returns while you read

You ask CLIO to check several files. It delegates one check and continues with the others. When the delegated check ends, CLIO records a background completion injection. The parent reads the result and can incorporate it into its answer.

Follow the task or child-agent details to inspect the returned check. If the result says verification is incomplete, the completion notice does not change that conclusion.

Expanded Wait for all tasks showing one collected result and the child agent's calculation — Open the returned task result to read the check. Turn ended describes the child agent's status; the calculation beneath it is the result.

Example: a scheduled check begins

You ask CLIO to inspect a job later. A Scheduled wake carries that saved instruction into the conversation when it is due. A Loop wake does the same for a repeated check. Open the injection to see the instruction, then inspect the following operations for what actually happened.

The row records supplied context. To change or stop future scheduled work, use the controls in Work; collapsing an injection only changes its display.

Scheduled wake expanded above the resulting saved report — Scheduled wake carries the saved instruction into the conversation. The following response and output show what happened after it arrived.

Goals, loops and scheduled workSet up continued work and inspect its controls.

Follow work in progressFind the operation and result behind a notification.

What CLIO remembersInspect retained information used in a conversation.