Skip to Content
Skilder

Read Activity

Follow skills, sessions, tool calls and connected clients in Activity.

Open Activity and you see what your roles actually did: which sessions ran, which tools got called, and what failed.

Before you start

  • At least one role has run once.

Open Activity

Select Activity under Monitoring in the sidebar. The Stats tab opens first, on the last 30 days.

Top skills by uses ranks your skills by use. A skill counts once when learned in a session, even if that session makes several tool calls. Test calls do not count. Pick 7d, 30d or All time above it.

Rank every skill

Select Skills to rank every skill over 7d, 30d or All time. Choose Uses to see which skills get used most, or Unpriced uses to see only the skills used without a value per use, biggest gap first. Filter by Role to narrow the ranking, and click Load more below the list to see the next 50 skills.

Under Unpriced uses, the uses of a skill deleted since stay counted, in one Unattributed row. It adds up with the other rows to the unpriced count under Pricing coverage on Value, and it has no skill to open.

Click a skill to open it, then set its value per use in its Value section.

Follow a session's story

Select Sessions to see agent work grouped by task, one row per episode. Click a row to open its timeline: every call in order, nested under the call that triggered it.

One session's timeline: Jordan Lee's task to triage the Zurich warehouse tickets, six successful calls in order: init_skilder, learning the Support Engineer role and its Triage a Ticket skill, a query, then the skill's triage.py script and an execute call. Every call after init_skilder is labeled with its role, and with its skill when it ran under one.

A session starts when a client calls init_skilder and carries its session id on every call that follows. While session tracking is on, a call that cannot be tied to a session is refused. With it off, calls join no session and show up only in Call Log.

Look at every call

Select Call Log for a flat, filterable list of every tool invocation: status, guardrail verdict, role, skill, duration, estimated tokens, and who called it. Search, or filter by status, tool, runtime, role or time range, to narrow in on a problem, then click a row to inspect its full input and output.

The Call Log tab filtered to the On-call Engineer role, listing six calls with one failed execute_script call marked in red.

See what your guardrails decided

Select Guardrails for the calls a guardrail judged, newest first, with the rule and its reason on each row. The counts above the list name each decision: click one to keep only those calls. A contributor or member sees the decisions on their own calls.

Read the value your skills returned

Select Value to see what your skill uses were worth over 7d, 30d or All time. Total value adds up the value per use of every priced use. Pricing coverage beside it splits your uses into priced and unpriced: unpriced uses count as zero, so the total reads low.

Check connected clients

Select Clients to see which MCP clients have connected to your workspace.

Check it worked

A row with a red status icon in Sessions or Call Log is a failed call. Click it to read the error in its input and output.

Admin: workspace-wide activity

A contributor or member only sees their own skills, sessions and calls. An admin sees everyone's, can filter Skills by User and Value by person, and gets the admin-only Runtimes tab. Session tracking is a workspace setting: an admin can turn it off, and calls still appear in Call Log with no sessions recorded.

On Value, an admin also gets Top users by value. Click a row to filter every figure on the tab to that person, and click it again or Show all users to clear the filter. Find a person searches everyone in the ranking, up to 50 people.

The person stays selected when you switch to Skills.

Activity names each person by their display name, or by their full email address when they have none.

Next

On this page