Skip to main content

​
What this helps you do

An annotation is reusable guidance attached to a file or folder. It can explain what the content is for, identify authoritative sources, set boundaries, or tell an agent how to carry out work in that location. When a chat works inside a folder, Collivo provides the applicable folder annotations to the agent as standing instructions on every turn. This makes an annotation especially useful for orienting an agent to a folder’s purpose without repeating the same task brief in every message. Annotations and Hide from AI solve different problems:

​
Before you start

You need edit permissions for the target file or folder.

​
Add or change an annotation

  1. In the Files pane, open the target file or folder’s More options menu .
  2. Select Annotate.
  3. Add a short description, task guidance, or both.
  4. Review any guidance shown from parent folders so your new annotation does not conflict with it. Inherited guidance is labelled with the folder it comes from.
  5. Optionally select Check draft to review your text before saving.
  6. Select Save.
Use the same action to update the annotation later. Select Delete in the annotation editor when the guidance no longer applies.

​
Check a draft before saving

Check draft reviews the text you have written for clarity and for whether it belongs at this level. It returns a list of findings. Select the ones you want to act on, then either:
  • Fix selected with AI — have Collivo rewrite the text using those findings. Use the accompanying box to say how you want them applied.
  • Copy detail prompts — copy the findings out and edit the text yourself.
Check draft looks only at the annotation in front of you. It does not compare it against the persona or the inherited folder guidance active in a chat — see Active instructions for that.
When you save, Collivo may warn you that the current chat has an active persona or has muted this folder’s instructions. The annotation is still saved, but it may not affect that chat until you review or un-mute the active instructions. Choose Review now to open Active instructions, or Save now, review later.

​
Use a folder annotation as a task brief

A folder annotation can turn a project folder into a resumable task area. It gives a new agent enough operating context to determine what the folder is for, understand how decisions should be made, and continue from the recorded state.
For the full setup workflow, see Task folders.
A useful task brief can define:
  • The outcome the work should produce.
  • The agent’s responsibilities and the decisions reserved for the user.
  • The approved scope and any actions that require confirmation.
  • The workflow phases and review points.
  • What the agent should read when a session begins.
  • Where progress, decisions, deliverables, and blockers are recorded.
  • Which documents are authoritative.
  • Where drafts and final outputs should be saved.
For a long-running task, the AI’s working pattern becomes: Enter the folder → read the task brief → inspect requirements and progress → summarize the current state → confirm the next action → continue the work → record progress For example, a client-delivery folder could carry guidance like this:
Act as the delivery lead for the work in this folder. Prepare the required outputs and bring decisions that affect scope or acceptance criteria to the project owner. At the beginning of a session, read requirements.md and progress-journal.md, summarize what has been completed, and propose the next action. Follow the project phases in order and update the journal after meaningful progress, decisions, or blockers.
Keep durable operating guidance in the annotation and changing task information in referenced files: This separation keeps the task brief stable while allowing the work itself to evolve.

​
Apply annotations to common folder types

​
Historical archive

Prevent old material from being mistaken for current guidance:
This folder is a historical archive. Use its contents only for historical comparison. Do not treat these files as current or authoritative; use the documents in Current policies for current guidance.

​
Template library

Protect master templates while still allowing agents to use them:
This folder contains approved templates. Do not edit the originals. When a template is needed, create a copy in the appropriate working folder and make changes to the copy.

​
Controlled source hierarchy

Explain which information takes precedence:
Treat documents in Approved as authoritative. Material in Research can provide supporting context but must not override approved content. Identify any assumptions when an approved source does not answer the question.

​
Repeatable production workflow

Orient the agent to a specific process:
This folder is used to prepare the quarterly board pack. Begin with the reporting checklist and the previous approved pack. Use the current-quarter workbook for figures, flag missing inputs, and save new drafts in Drafts.

​
Write effective guidance

Keep annotations specific to the item and useful across multiple tasks:
  • State which source is authoritative.
  • Define the agent’s role and decision boundaries.
  • Name the intended audience, workflow, and output standard.
  • Explain important exclusions and actions the agent must not take.
  • Reference the files that hold current requirements or progress.
  • Remove obsolete instructions promptly.
Keep broad style or role preferences in a persona. Use annotations for guidance tied to project content or location.

​
Understand folder guidance

The annotation editor shows relevant guidance from parent folders so you can understand the surrounding context. When an agent works on a target inside that hierarchy, the applicable active guidance is combined in order. Use the current folder’s annotation for the instructions specific to the task performed there. Direct annotations on files also travel with the file when the AI reads it, helping the agent interpret the content correctly.

​
Hide content from AI discovery

Agents can explore a project to find useful context. That exploration can also surface outdated, duplicated, incomplete, or purpose-specific content. Hide from AI keeps such content out of normal agent search and listing without moving or deleting it. To hide a file or folder, open its More options menu and select Hide from AI. Use Unhide from AI when it should become discoverable again. Common uses include:
  • Historical archives: keep superseded policies, expired contracts, and old plans out of current answers.
  • Template libraries: prevent master templates from being selected or edited during routine work.
  • Drafts and experiments: stop abandoned approaches, rough notes, or incomplete analysis from influencing approved work.
  • Generated output and backups: prevent old exports, intermediate files, or previous agent output from being treated as primary sources.
  • Large reference collections: keep routine exploration focused while making individual references available when needed.
  • Sample or test content: reduce the chance that fictional records or demonstrations are mixed with operational information.
  • Human-managed records: keep administrative or evidentiary material available to authorized people but outside ordinary agent exploration.

​
Make hidden content available intentionally

Hiding controls discovery, not permissions. You can still provide a hidden item to an agent using an @ reference, the Context selector, or another supported reference. Once the agent is made aware of the item, it can use it like other content you have permission to use. The item remains hidden from future autonomous exploration unless you select Unhide from AI.
Hide from AI is not a security control. People who already have permission can still see hidden items in the Files pane and provide them to an agent. Change the item’s sharing permissions when someone must not be able to access it.

​
Expected result

The folder gives agents a repeatable operating brief, project content is interpreted according to its intended purpose, and irrelevant material stays outside normal AI discovery until a user chooses to provide it.

​
Troubleshooting

  • The agent does not follow the task brief: confirm that the chat is working in the intended folder, then open Active instructions to check the guidance is Active rather than muted.
  • Guidance conflicts: simplify the current annotation and align it with the applicable parent guidance shown in the editor.
  • Archived material is treated as current: name the authoritative replacement directly in the archive annotation and consider hiding the archive from AI discovery.
  • The agent can read a hidden file after you referenced it: this is expected. Hiding prevents autonomous discovery, not intentional use by someone with permission.
  • Content must not be available to someone: change its sharing permissions instead of relying on Hide from AI.
  • Annotate is disabled: confirm that you have edit permissions and that the target is not a shortcut.