The packaged admin skills

Seven procedures ship with the package. The agent fetches one with Get Skill and follows it. Each has a fixed output shape, so two runs are comparable. Skills propose; publishing still goes through Activate Draft with your approval.

These seven belong to the admin agent, and the permission model gates them. Studio neither lists nor edits them. The Skills page shows a different set: the skills your org writes for its own agents. A ContextWorks Admin can still reword one of the seven, or add an eighth, through Save Draft with surface=admin and then Activate Draft.

Composing

• document_to_briefs — turns a methodology deck, enablement guide, or policy page into typed brief drafts, one per audience. It presents a review table first, then drafts. Where material is sensitive, it says so and recommends a narrower domain.
• compose_entity — drafts an entity’s text: the catalog description, the instruction, and synonyms drawn from the org’s own vocabulary. Shows before and after, then saves the draft.

Reviewing

• review_pending_changes — finds every draft ahead of its served version and walks each through the publish flow: preview, your yes, commit with a comment. Ends with a ledger of each one — published, held, or needs Studio.
• config_conflict_audit — finds contradictions, redundancy, and staleness across instructions, briefs, and domain meaning. Reports findings with locations and excerpts; changes nothing.
• compare_artifact — puts two artifacts of one library side by side: entity vs entity on one object, brief vs brief, domain vs domain, or a copy against its origin. Each difference gets a verdict — aligned, deliberately divergent, contradiction, or silent drift — with one recommended action per problem row.

Operating

• health_review — the weekly check-in: trends, cost movers, quality dips, and issues, read against the change log so a movement gets a suspected cause. Returns ranked findings, each with a next action.
• diagnose_payload — traces “the agent said something wrong” or “didn’t know X” to one cause: a permission strip, a failed element, an empty field, an artifact that is not active or not in the domain, or a recent change. Returns the diagnosis, the evidence, and one fix.

What’s next

• The tools