Skip to main content
Knowledge and Skills solve different problems:
  • Knowledge holds approved source material, brand facts, reference context, and grounded artifacts.
  • Skills holds reusable working procedures and the conditions under which agents should use them.
Open Knowledge or Skills and confirm the active workspace.

​
Decide what you are adding

Put changing performance numbers in Analytics or their owning surface. Put a repeatable method for evaluating those numbers in a Skill.

​
Follow the context lifecycle

1

Establish authority

Confirm source owner, effective date, approved use, and whether a newer source conflicts. Remove secrets and personal data the workspace should not retain.
2

Build or upload

Use Build from website for the canonical public site, or route approved files through Sources. Wait for a terminal state and verify the resulting Brand kit, Knowledge, or Skill badge.
3

Inspect grounding

Review the Standing index, open Recent artifacts, and inspect the sources named by Grounded in N. Distinguish sourced fact from synthesis and recommendation.
4

Govern procedures

Review Brand foundation, active Skills, Ready to promote, proposed and archived items, and learning evidence. Read every imported procedure before activation.
5

Verify in a fresh task

Test one fact or method in a new task, require the named source or procedure, then run a nearby negative test that should not trigger the Skill.

​
Go deeper

Knowledge sources

Build from a website, route files, change destinations, inspect grounding, and test retrieval.

Skills

Audit active procedures, promote candidates, import approved Skills, and test both sides of the trigger.

​
Important boundaries

  • The Standing index is compact context, not the entire knowledge base.
  • A website build reflects the site when research ran; rebuild after material changes.
  • Moving a source out of Brand kit does not guarantee previously extracted brand details disappear.
  • Images and fonts behave as brand assets rather than ordinary factual documents.
  • Brand foundation is always on and is not deleted like an ordinary Skill.
  • Imported Skills become active after import, so the preview is an approval boundary.
  • Usage and win rate are review signals; Outcomes pending is not 0% performance.

​
Completion check

  • Every important source has an owner, date, authority, and intentional route.
  • Website and file work reached a terminal state.
  • Artifact claims can be traced to the named sources.
  • Active Skills have a specific trigger, scope, steps, branches, approvals, stop conditions, and final check.
  • Imported or promoted Skills were reviewed for overlap, secrets, unsupported authority, and stale assumptions.
  • Fresh positive and negative tests confirmed retrieval and Skill scope.

​
If context is wrong

Check source state, route, effective date, and conflicts before editing the prompt. Replace stale evidence, reroute incorrect material, and retest in a new task. When a Skill is ignored or overused, correct When to use, exclusions, scope, or status rather than duplicating the procedure.

​
Next step

Use context in chat

Attach the exact record and verify how the agent uses it.

Apply context to creative

Carry approved evidence and methods into Briefs and Studio.
Last modified on August 9, 2026