You installed skills and MCP servers. Which ones does Codex actually use, and do they work? The Capabilities view of Usage Insights for Codex answers that from your session files, one table at a time.
Open Capabilities
Press Capabilities in the list of views. The row at the top counts Skills, Connectors (MCP servers), Tools used, Subagents / research, Slash commands and New this range: everything used for the first time in the range you picked.
Skills, and who started them
Skills lists every skill with its uses, split into You and Codex:
- You: you named it with
$skill-name. - Codex: it loaded the skill on its own because a task matched its description.
A skill Codex never picks on its own usually has a description that does not say when to use it. A skill nobody used in 30 days is one to remove.
MCP servers, tool by tool
Connectors gives each MCP server a card: its calls, sessions and last use, then each of its tools with calls and success rate.
A tool with a low success rate is the first thing to look at when a server feels unreliable. A server with one tool in use out of twenty is a candidate for Only these tools in its config.
Built-in tools
Tools used lists Codex's own tools, such as exec_command for commands and apply_patch for edits, with calls, Success, Rejected (declined by you or blocked) and a 30-day sparkline.
Subagents and automation
Subagents / research lists each agent Codex handed work to, with its runs, average time and tool calls per run, read from the subagent's own session file. Under Automation, codex exec runs counts the sessions your scripts and CI started.
Plugins and Slash commands sit beside the skills, for the plugins whose skills you used and the commands you typed with a slash.
Next: is Codex working well?, turn times, errors and where the time goes.