Keep Knowledge Aligned
When knowledge changes in one place, the cards that depend on it somewhere else can fall out of date without anyone noticing. This skill watches a source for new or updated content, finds the cards in a target collection that are affected, makes targeted updates, and drafts new cards where nothing covers the change yet. Common setups include Product to Sales, Policy to Support, and Engineering to Customer Success. Everything it writes is grounded in documented content, and anything it can't support is flagged for a person to fill in rather than guessed.
Every team sets this up differently, so rather than starting from a fixed template, you can use the prompt below to have Operator Mode build the skill with you. It asks one question at a time about your source, target, triggers, and approval rules, then drafts the skill from your answers for you to review before saving.
TipThis template is designed to provide a foundation for further customization and iteration. It is not intended to be an out-of-the-box solution. Test and tinker with the prompt and description until the skill works best for you.
Before you start: check access and connections
Before you build the skill, confirm the following:
- Source access: The Knowledge Agent needs read access to the collection where changes start. If the source is an external tool, it must be connected to Guru as a Source or as an MCP integration before the skill can read it.
- Target access: The Knowledge Agent needs permission to create drafts in, and publish to, the collection it keeps aligned. If it can only create drafts, answer question 8 with "leave as drafts."
- Summary delivery: To post the run summary to a Slack channel, the Slack MCP must be connected and authenticated in Guru. Check under Apps and Integrations before moving to Step 1.
Step 1: Open Operator Mode and paste this prompt
Open Operator Mode on the Knowledge Agent that will run the alignment, then paste the prompt below. Operator Mode asks each question in turn and shows you the full skill before saving it.
I want to build a skill that keeps related knowledge aligned across my workspace.
Whenever knowledge is created or changed in one place (the source), the skill should
find the cards in another place (the target) that are affected, update them, and
draft new cards where nothing covers the change yet. I don't want to build this from
scratch. Instead, please ask me the following questions one at a time, wait for my
answer before moving to the next, and then use my answers to draft the skill for me
to review before saving.
Here are the questions to ask me:
1. Where does the triggering knowledge live? Name the source collection, and any
folders within it I want to limit this to, or specify the external source details
where new or updated content will be added.
2. What content in that source should trigger an alignment check? For example: every
card, only cards built from a specific card template or title pattern (such as
"Release Notes" or "Policy Update"), only cards with a certain tag, or for an
external source, a specific item type, project, or label. Should newly created
content, updated content, or both count as triggers?
3. How far back should the skill look each time it runs? If it runs on an
event-based automation, such as when a card is published or updated, it should
look back 10 minutes. If it runs on a scheduled automation, it should look back
to the previous scheduled run (for example, a weekly scheduled automation looks
back 7 days).
4. Where should the skill look for cards to keep aligned? Name the target collection
and/or folders.
5. What kind of information in the target cards needs to stay aligned with the
source? For example: feature availability, pricing, eligibility rules, process
steps, owners, or talk tracks.
6. If a source change affects a topic that no existing target card covers, should the
skill create a new card draft for it? If yes, which existing card should new
drafts use as a model for format and structure?
7. Are there any reference cards the skill should cross-check to decide when a new
card is needed? For example: if a source change overlaps with a topic covered in
a competitor profile, create a new battlecard draft. If so, tell me which cards or
folders to check and which existing card to use as the format model.
8. Should updates to existing target cards be published automatically, or left as
drafts for review? New cards are always left as unpublished drafts.
9. Who should receive a summary after each run? Give the email address of each
person who should be notified, or name a Slack channel to post the summary in.
To post to Slack, the Slack MCP must already be connected to Guru.
Once you have my answers to all nine questions, draft the skill. The skill should
include:
- A Skill Description that says when to use the skill and when not to.
- A role statement scoped to keeping my target collection aligned with my source.
- A workflow that:
- confirms the source, target, and date window before doing anything else
- finds qualifying source content, excluding drafts or unpublished items, and
stops with a "no alignment needed" report if none qualify, without searching,
editing, or notifying
- searches the target for affected cards using each change's name and key
concepts, and reads promising matches in full before deciding
- chooses the destination folder for new drafts by reviewing existing folders
and their cards
- reviews at least two existing cards in the destination folder and matches
their style and structure
- chooses one outcome per change: update an existing card, create a new card
draft, no change, or needs review
- makes the smallest edit that keeps each card accurate, and references or links
to the source content where the card format allows it
- verifies each update or draft was saved and inspects the result
- sends the summary to the email addresses or Slack channel from question 9
- Formatting rules: use Guru's native editor blocks (headings, paragraphs, lists,
tables) by default. Use a Markdown block only when the source calls for it, such
as a Mermaid diagram for a multi-step process. Never rewrite or convert a whole
card just to change one part of it.
- A summary format with bold section headers that list card titles as links, with
"None" under any section that has no items:
- "New Card Drafts Needing Your Approval"
- "Card Updates Automatically Published" if my answer to question 8 is to
publish automatically, or "Card Update Drafts Needing Your Approval" if
updates stay as drafts
- Items left unchanged or marked Needs review are listed last, each with a
one-line reason.
- Guardrails that include all of the following:
- Work only from what is documented in the source content and the existing target
and reference cards. Never invent or assume facts, figures, dates, names,
customer examples, competitor details, or claims to fill a section.
- If a section of a new draft cannot be supported by documented content, keep the
section and mark it inline with "Needs input:" followed by what is missing.
Do not write around the gap with general-sounding text.
- If the source is ambiguous or conflicts with another documented source, mark the
item Needs review and do not edit or publish anything for it.
- Never edit cards outside the named target collection or folders, and never use
content outside the named source as the basis for a change.
- Never create new folders or drafts at the collection root, and never create a
card that duplicates one that already covers the topic.
- Never overwrite a card wholesale when a targeted edit keeps its useful content.
- Never publish a new card. New cards stay as drafts for a person to approve.
- Never report an action as complete unless it was confirmed. If a publish, draft,
link, or notification fails, say so in the summary.
Before saving, show me the full draft and ask me to confirm. After saving, tell me
to add any format model cards from questions 6 and 7 to the skill as reference
files myself: open the skill instructions, type "@", then type the card title and
select the card.Step 2: Add reference cards and attach the skill to an automation
Add your format model cards. Operator Mode can't attach reference files for you. After the skill is saved, open its instructions, type @, then type the title of each format model card you named in questions 6 and 7 and select it. See Adding references to a skill for more detail.
Attach the skill to an automation. Set up a Knowledge Agent automation to run the skill. You can trigger it from an event, such as a card being published or updated in the source, or run it on a recurring schedule. Make sure the lookback window you gave in question 3 matches the trigger type. A mismatch either skips changes or processes the same change twice.
Updated about 1 hour ago
