Skip to main content
Skills are reusable working procedures for the Marketing Agent and specialists. A Skill defines when a method applies, which agents can use it, what steps to follow, and where human review or a stop condition belongs. Open Skills and confirm the active workspace.
Skills header explaining on-demand tactical chunks, outcome scoring, promotion, and reversible edits, with an Import SKILL.md button

Skills explains that procedures are used when relevant and evaluated from outcomes, with the import control at the top of the surface.

​
Read the Skills surface

The page can organize procedures and learning into several zones:
  • Brand foundation
  • Ready to promote
  • Proposed
  • Active Skill library
  • Archived
  • What the agents learned
These zones are not equivalent. A learned pattern or proposed candidate is not active workspace policy until it is reviewed and promoted or saved as an active Skill.

​
Review Brand foundation

Brand foundation is always on. It supplies foundational brand procedure and context to agents.
  • Its identity and scope are fixed.
  • It cannot be removed like a normal editable Skill.
  • Correct the underlying Brand Profile or approved brand sources when its behavior is wrong.
  • Do not create a duplicate Skill merely to override a stale foundational fact.

​
Audit the active library

The active table can show:
  • Skill
  • Scope
  • Win rate
  • Used
  • Last used
  • Pinned
Open a Skill and review:
  • Name
  • Scope
  • Status
  • Description
  • When to use
  • Markdown Body

​
Interpret usage signals

  • Outcomes pending means the Skill has usage without enough resolved outcomes. It does not mean 0% success.
  • A low or underperforming signal is a review prompt, not an automatic delete command.
  • Pinned protects an important Skill from automatic archival behavior; it does not increase authority or force every agent to use it.
  • Rare usage can indicate a narrow, healthy trigger or an unclear trigger. Read the actual requests before deciding.

​
Write an executable Skill

A useful Skill body contains:
  1. The exact situation in which it should run.
  2. Explicit situations in which it should not run.
  3. Required inputs and source authority.
  4. Ordered steps.
  5. Decision branches.
  6. Approval or permission boundaries.
  7. Stop and escalation conditions.
  8. A completion check based on visible customer outcomes.
Use Scope to choose from the specialists offered by the editor or all agents. Do not name a specialist the UI does not offer, and do not use an all-agent scope for a provider-specific operating method without a reason. Select Save only after When to use and the body agree. Use Delete only after identifying the exact editable Skill; use archive or replacement when preserving its history matters.

​
Review Ready to promote

Ready to promote contains candidates that need a human decision.
1

Read the evidence

Open the candidate and inspect the outcomes or repeated pattern supporting it.
2

Check for overlap

Compare its purpose and trigger with Brand foundation, active Skills, proposed items, and archived history.
3

Choose Edit when needed

Tighten the name, scope, trigger, ordered steps, stop conditions, or final verification.
4

Choose Promote deliberately

Promote only when the procedure is distinct, source-safe, and appropriate for repeated workspace use.
5

Choose Dismiss for an unsuitable candidate

Dismiss removes it from the current review. A relevant pattern can be suggested again after later evidence.
Review Proposed, Archived, and What the agents learned during the same governance pass so a promotion does not recreate an existing method.

​
Import approved Skills

Select Import SKILL.md. Workspaces on the newer label can show Import skills for the same import flow. The dialog supports a file/drop route and pasted content.
1

Add the source

Drop the approved Markdown file or paste its content, then select Add to preview.
2

Review the preview table

Confirm Source, Name, Scope, and Status for every entry. Remove duplicates, overly broad scopes, and entries without usable triggers.
3

Read the full body

Check for secrets, environment assumptions, unsupported account changes, missing human review, stale provider labels, and references that do not exist in Marketer.
4

Select Import

Confirm the reviewed count. Imported Skills become active immediately, so the preview is the approval boundary—not a staging area for unread procedures.
5

Reopen each imported Skill

Verify its name, scope, status, description, trigger, and body survived the import as intended.
Imported or promoted Skills can influence future tasks across their scope immediately. Never activate a procedure whose authority, approval boundary, and stop conditions you have not read.

​
Test both sides of the trigger

For every new or materially changed Skill:
  1. Start a fresh task that clearly should use it.
  2. Ask the agent to name the procedure it followed and show the customer-visible steps.
  3. Confirm it uses the right sources, branches, approval, and completion check.
  4. Start a nearby request that should not use the Skill.
  5. Confirm the agent does not force the procedure into that request.
  6. Tighten When to use, exclusions, scope, or stop conditions if either test fails.
A positive-only test cannot detect an overbroad trigger.

​
Completion check

  • Brand foundation is understood and not duplicated.
  • Every active Skill has a specific scope and trigger.
  • Body steps, branches, approvals, stop conditions, and final verification are explicit.
  • Usage, win rate, outcomes pending, underperformance, and pin state are interpreted correctly.
  • Promoted candidates are distinct and supported by evidence.
  • Imported Skills were fully read before Import.
  • One positive and one negative fresh-task test passed.

​
Troubleshooting

Confirm Status, Scope, and When to use match the real request language. Test in a fresh task without relying on an older conversation’s context.
Add explicit exclusions and stop conditions to When to use and the body, narrow Scope, then rerun nearby negative-trigger requests.
Check whether the Skill shows Outcomes pending or has too little resolved usage. Review individual outcomes rather than treating a missing percentage as failure.
Keep one authoritative procedure. Merge useful evidence or steps into it, then dismiss, archive, or delete the duplicate as appropriate.
Reopen it and compare the imported name, scope, status, trigger, and body with the approved source. Correct unsupported assumptions, save, and rerun both trigger tests.

​
Continue

Manage Knowledge sources

Keep facts, brand material, and procedures in the correct route.

Verify in a real task

Test the procedure with explicit context and visible completion criteria.
Last modified on August 9, 2026

Related topics

Knowledge and SkillsCreate a reusable SkillNext stepsChat and contextProduce a creative brief