Skip to main content
Cedros

Enabling and customizing assistant skills

Find and enable assistant skills, tailor their instructions, create a custom skill, and restore the original wording.

Skills give the Cedros assistant reusable instructions for particular kinds of work. Use them to set a consistent approach to tasks such as preparing customer replies, reviewing content, or following a repeatable workflow.

You can enable existing skills, tailor their instructions, or add a custom skill for your site. Skills guide the assistant's work; its tool access and your account permissions still apply. See Controlling what the assistant can do when you need to change those limits.

Find a skill

Open Settings → Assistant → Skills. Use Search to find a name, capability, or trigger phrase, and Filter to narrow the list.

  • Service skills are packaged skills available on your site.
  • Workflow skills describe work involving a sequence of skills.
  • Custom skills are instructions you add yourself.

Use Disabled to find skills that are switched off, or Customized to find saved instruction overrides. A Default badge means the skill uses its original instructions; Customized means it has a saved override.

If the Skills tab is missing, ask your administrator to check assistant memory access for the site. If you see a provider setup notice, follow Choosing AI providers and models before continuing.

Read a skill before enabling it

Open the skill's row and review its Instructions. Then check the reference cards:

  • When to use and When not to use describe situations the skill should match or skip.
  • Permissions, when shown, identifies declared behaviors such as writing files, writing memory, running scripts, or publishing externally.
  • Capabilities and Allowed jobs list the operations and background jobs declared for the skill.
  • Workflow steps, when shown, lists the skills in a workflow.

These cards are reference information, not switches you edit on this page. A risk badge helps you recognize sensitive work; choosing or seeing a risk level does not itself block an action or grant permission. An absent Permissions card means no such badges are listed, not that every possible use is harmless.

Turn a skill on or off

  1. Choose Back to Skills if you are in the detail view.
  2. Use the skill's Status toggle to enable or disable it.
  3. Wait for the success message and check the resulting status.

The toggle saves immediately; there is no separate publish step. Disabled skills remain in the list and can be enabled again later. Enabling a skill makes it available for relevant requests; it does not start a task by itself.

Turning a skill off does not undo earlier work. If you need to stop work already in progress, use the chat stop control and review queued requests and any affected items as well.

Customize the instructions

  1. Open the skill and read its current instructions before editing.
  2. Edit Skill instructions. Keep the parts you still need and add your site's wording, boundaries, and examples.
  3. Choose Save changes and wait for the saved message.
  4. Reopen the skill to confirm the text and Customized badge.

Your saved text replaces the default instructions for that skill. It is not an extra note appended to the original, so preserve any steps that still matter. There is no separate publication step after saving.

Instruction edits do not autosave. Back to Skills discards text you have not saved. Copy unfinished wording elsewhere if you want to return to it later.

Useful instructions explain:

  • When the skill should apply and when it should be skipped.
  • Which sources to use and which facts must be preserved.
  • The steps to follow and the format of the result.
  • Whether to answer in chat, prepare a draft, or perform an explicitly requested action.
  • What to do when information is missing, contradictory, or unavailable.

Keep the wording focused. Writing “publish automatically” into a skill does not supply a missing permission, and asking it to avoid publishing does not change the site's configured access levels.

Add a custom skill

  1. Choose Add skill from the list.
  2. Enter a clear, distinct Name and the required Instructions.
  3. Add a Description that explains when the assistant should use it.
  4. Optionally choose Low, Medium, or High under Risk level; leave No risk level if you do not want to assign one.
  5. Choose Add skill, wait for the success message, and find the new Custom row.

New custom skills are enabled when created. Review the instructions before adding one, and switch it off afterward if you are not ready for it to be used. The detail editor lets you revise the instructions; the name, description, and risk level are not editable there.

For example, create a skill named Customer reply style, with the description “Prepare concise replies to customer enquiries using supplied facts.” Its instructions could be:

Use this skill when asked to draft a customer reply. Read the supplied enquiry and confirmed facts. Write a courteous reply in plain language, under 150 words. Preserve prices, dates, and policy terms exactly. Do not invent a promise, offer, or deadline. Ask for missing information that would change the answer. Return the reply in chat for review; do not send it or change customer records.

Adapt the example to your actual process. A custom instruction does not install a new integration or create a tool the assistant does not already have.

Check that the skill helps

First confirm that the instructions saved and the skill is enabled. If you want to try it in chat, allow for normal AI usage charges and use a small, non-sensitive example.

  1. Name the skill in your request and describe the task. For example: “Use Customer reply style to draft a response to the enquiry below. Show it here without sending.”
  2. Provide the source material the skill needs.
  3. Compare the result with the saved instructions: required facts, length, tone, and intended action.
  4. If the result differs, clarify the instruction, save it, and try a focused follow-up.

Include a case with a missing detail, such as an unconfirmed date, to check that the assistant asks instead of guessing. A successful save confirms that the instructions were stored; it does not demonstrate how every request will behave.

See Writing useful requests for the assistant for help giving the skill enough context, and Getting started with the Cedros assistant for chat controls.

Restore the original instructions

  1. Open a skill with a Customized badge.
  2. Copy the current wording if you may want it again.
  3. Choose Reset to default and wait for the restoration message. Check that the original instructions return and the Customized badge disappears.

Reset applies immediately. For a packaged skill, it restores the packaged instructions. For a custom skill, it restores the instructions entered when the skill was created. It does not delete the skill or change its enabled status.

The reset button is unavailable when no saved instruction override exists. To discard only an unsaved edit, return to the list without saving instead.

Resolve common problems

  • No skills match this view: clear Search and choose All skills. No skills installed means the catalog itself is empty.
  • Editing controls are disabled: wait for loading to finish, then ask an administrator for permission to edit shared assistant settings if they remain read-only.
  • Add skill will not submit: enter both a name and instructions. If a name is already taken, use a distinct name; changing capitalization or spacing alone may not make it unique.
  • Save changes is disabled or edits disappeared: Save changes becomes available when the text differs from the saved instructions. Leaving through Back to Skills discards unsaved text.
  • An enabled skill is not helping: check its saved instructions and When to use guidance, name it in a focused request, and verify the required provider and tool access. Do not enable broader permissions simply to make the skill run.

If adding or saving reports an error, read the message and reopen the list or skill to check what was stored before submitting again.