Dashboard Guide

Skills

Which skills your sessions load, what calls them, and what they set off while active.

What the Skills pages show

Skills in the sidebar covers every time Claude Code loaded a skill with its Skill tool, in the selected date range. It answers:

  • Which skills do my sessions actually use, and how often?
  • How much work does each one set off?
  • What invokes a given skill, and what does it call while it's active?

Skills are grouped by name, as Claude Code reports it, for example superpowers:brainstorming.


Skills list

Summary cards

CardWhat it shows
Total InvocationsSkill loads in the period, across all skills
Unique SkillsHow many different skills were loaded
Avg Tools / InvocationTool calls triggered, divided by invocations
Cross-Session ToolsSee the note below

Skill Breakdown

One row per skill, most-invoked first. Click a skill's name to open its detail page. A description appears beside the name for skills Trenchcoat recognises.

ColumnWhat it shows
SkillThe skill's name. Links to the detail page
InvocationsTimes the skill was loaded
Tools TriggeredTool calls started while the skill was active
Avg Tools / InvocationThe two above, divided, to one decimal
Cross-Session ToolsSee the note below

Cross-Session Tools always reads -- or 0. It counts tool calls in other sessions started by a skill, and relies on a field the plugin stopped writing, because subagents run inside their parent's session rather than starting their own. The detail page leaves it out.

What "triggered" means

A tool call counts toward a skill when it's made while that skill is the active context in the same turn. That is how the plugin attributes calls to a skill: it's a window, not proof that the skill caused the call. A skill loaded inside a subagent is tracked separately for that subagent, so one subagent's calls aren't credited to another's skill.


Skill detail

The detail page covers one skill over the selected range.

Summary cards

Invocations, Tools Triggered, Avg Tools / Invocation and Sessions. The first three use the same definitions as the list, so they match the row you opened the page from.

Relationships

A picture of what the skill is connected to. The skill sits in the middle.

  • Left: Invoked by. What loaded it:

    • the main conversation
    • a subagent type, if the skill was loaded inside a subagent
    • another skill, if one was already active when this one was loaded

    If both a subagent and a skill apply, the subagent is credited.

  • Right: what it set off while active, in labelled groups:

    • Launches: subagent types it started
    • Skills: skills loaded while it was active
    • MCP servers and Tools it called

The number beside each item is invocations for callers, launches and skills, and calls for MCP servers and tools. A thicker line means more. When a skill is loaded while an earlier run of itself is still active, the count appears under its name in the middle ("invoked inside itself 2×") and as a row marked "(itself)" in both lists.

Reading it works the same as on the Agents page:

  • Hover or click an item to highlight its line and see its figures in the panel on the right.
  • Subagent types, skills and tools link to their own pages, so you can follow a chain: from a skill, to the subagents it launches, and on to what they use. MCP servers don't link, because a server has no page of its own.
  • The chips above the picture show and hide whole groups, and long groups fold into a "+N more" item you can open.

The Tools Triggered card and the picture count differently, so their totals won't match. The card counts every call started while the skill was active, the same way the list does, including loading another skill and launching a subagent. The picture counts completed calls, and shows skills and subagent launches in their own groups rather than as tool calls. A call that never reported finishing also counts only in the card. The picture counts completions because only they carry a call's duration, result size and, for a launch, which subagent type it started.

An Agent item under Tools, rather than a subagent type under Launches, is a launch whose subagent wasn't identified, so its type is unknown.

What it leaves out, deliberately: cost, since tool calls are never priced individually, and error counts, since tool failures aren't yet captured reliably enough to count.

Charts

Invocations per day and Tools triggered per day.


Filtering

  • Date range: the picker at the top right. Everything on both pages uses it.
  • Machine: the global machine filter. Opening a skill, and following a link from one detail page to another, keeps both filters.

On this page