Endpoint AI

Skill load analytics and enablement

Skill load analytics and enablement

Harriet tracks skill loads — when a skill is brought into an active execution context in chat, workflows, Harriet Desktop, or Claude Desktop (via observable lazy loads). Loads indicate engagement; they do not measure whether a task completed successfully.

Who sees what

  • Workers see company-wide aggregate popularity on skill cards and Skills for you in My AI — popular requestable skills with one-click Request access, plus their own recent load history. They do not see which colleagues loaded a skill.
  • Provisioner admins with audit access see dashboards, the Skill Analytics page (Admin nav), per-skill usage tabs, named loader breakdowns, enablement recommendations, and CSV export.

My AI: skills for you

On My AI, the Skills for you card lists skills your organization uses that you do not have yet. Popular skills (with load counts) appear first; other requestable skills are shown as new to your org. Use Request access (or Get access when auto-approval is enabled). Pending requests show Request pending.

Skill Analytics (admins)

Open Skill Analytics from the Admin nav (or Analytics on the Skills list, or View analytics on the Dashboard). Filter by date range, surface, skill, team, or person. Switch the chart between loads by surface or by skill, and the breakdown table between skill, person, team, or surface. Click a row to open the skill editor, person/team detail, or audit log. Export the current breakdown as CSV.

Dashboard enablement plan

On the Provisioner Dashboard, the What should we do next? section turns load patterns into explainable recommendations — for example skills that are assigned but never loaded, declining usage, or many pending access requests. Each row links to the relevant skill or workflow so you can assign, share, or review.

Known limits

  • Claude Desktop package skills loaded only on the client (without an MCP round-trip) are not visible in analytics.
  • Group and profile breakdowns use current membership, not historical assignment at load time.
  • Anonymous widget loads are counted but do not contribute to unique loader counts.

Related

Use Harriet in your organisation for searchable help, AI assistance, and your company knowledge base.

Log in to Harriet