What's new

Minor UX/UI improvements

The free trial is now 3 days of use, 7 calendar days at most. Week and Month, editing and the other actions need a licence after that.

Custom types is now called Custom Data Types, in Settings and in its window.

Skills ends on the same Total tokens row as every other tab. Click it to switch to No-skill tokens, in any sort.

The menu bar bolt turns orange only while a chat's context keeps growing. Green cards, like an expired cache, no longer colour it.

Custom tab: add your own kinds of files with a path mask in Settings, with a licence. Directives, personas, workflows, scripts, wiki pages – each type opens into its files, with tokens and launches for the period.

Click a file to see where it runs, which skills, commands and agents it was used with, and what was read next to it. Right-click copies the command that runs it.

Files you attach with @ count as reads now.

A card in the bell tells you when skills nobody used this month still cost tokens on every request: pause them all or open the list.

Pause is offered only where it works – Claude Code.

Long names scroll on hover.

Rate SkillKeeper when your trial ends and get a month of free use.

Gift licence keys can carry an end date.

Indexing no longer gets stuck while a Cursor or Codex session is running. A long log is read on from where the last scan stopped – a 200 MB Codex log took 33 s on every change, now under a millisecond.

Cursor, Codex and Gemini show "Indexing N of M" during the first scan.

Settings → the bug icon, bottom left, copies diagnostics to paste into a support email.

Bug fixes.

Works on Windows and Linux through the Claude Code mod – one licence for the app and the mod.

The bell in the panel footer shows what needs you: a chat waiting for your reply, an agent spending tokens fast, a skill description that's too long. Open the chat where it runs – Antigravity, VS Code or its Terminal tab – or mute what you don't need.

Every row on the main screen opens a popover now: what it is, where it's stored, tokens per call and where it ran.

While Claude Code works, the app uses 5% CPU instead of 13%, with peaks of 30% instead of 90%.

When a chat's context gets expensive, the menu bar bolt turns yellow.

Works with Codex, Cursor and Gemini CLI – pick yours in Settings.

Codex and Gemini CLI show tokens per skill, like Claude Code.

Cursor shows skill counts only: it keeps token counts on its servers.

Skills your tool loads on its own show up without adding a folder.

UX improvements.

All – every skill, agent, tool, MCP server, command and memory file in one list, sorted by what it cost.

More sources behind the third tab: MCP, Commands, Memory, Hooks and Plugins, next to Tools.

5h – a new timeframe for the last five hours, with the 5-hour limit next to it.

Hover a bar in the chart to see what went into that half-hour, hour or day – the list counts only that slice.

Sessions – a new tab in the dropdown: tokens by conversation, with a click to open it.

Context – when a chat nears its model's limit, it shows under Total tokens with its size, and the menu bar bolt turns orange. Click it to open the chat.

Faster start – the current timeframe loads first, older history fills in behind it.

Minor UX/UI improvements, including a cleaner number column and a steadier chart.

See where your tokens actually go – switch the list between Skills, Agents and Tools. Each skill now counts only the tokens of its own turn, and the rest shows up as a separate No skill row.

Agents – every subagent type with its launches and tokens, next to the main session.

Tools – tokens split across Bash, Read, Edit and each MCP server.

Skills read directly from their SKILL.md now count as runs, and skills loaded at session start get their share of tokens.

Minor UX/UI improvements, including a floating panel that no longer slips under the menu bar.

Pause one of two look-alike skills to resolve a clash – the warning clears on the other, while the paused skill stays grey as a reminder.

Search the skills.sh store from the same field – tap the store button next to the X, or just stop typing, to see results by installs. One click copies the install command.

Right-click a skill to copy its slash command – /name for a skill, /router subskill for a subskill. Loop is now Option + right-click.

Pin by dragging – pull the popup away from the menu bar to turn it into a floating window that stays on top on every Space. The pin button is gone.

Archived Skills now keeps archived skills grouped, and Restore puts them back in their original group.

Minor UX/UI improvements, including a smoother sort change.

Minor UX/UI improvements

Minor UX/UI improvements

Search by skill name – swipe up on the list, past its top edge, and the search field slides in. Type to filter, Escape or the X to hide it

Minor bug fixes

Rescans the moment something actually happens – file-system events instead of a fixed timer, so a skill call shows up within a couple of seconds, not up to 30 seconds later

Scan results are cached between launches – relaunching takes about a second instead of re-reading every transcript

A growing transcript is read from where the last scan stopped, not from the top – an active Claude Code session no longer costs CPU on every rescan

Point SkillKeeper at your project folder too – that's usually where a project keeps its own skills, alongside the default ~/.claude/skills

Reads every Claude Code transcript (~/.claude/projects/**/*.jsonl) – works for claude-desktop, claude-vscode, and the plain terminal SDK/CLI

Counts direct Skill tool calls, and detects subskills invoked through the router→Read pattern (a router opens, then reads a subskill's SKILL.md right after)

Caches by file mtime – untouched transcripts aren't re-read on the next scan, so re-scanning after the first full pass is nearly free

Three timeframes – 24 hours / Week / Month – switch instantly, no new disk scan

Sort by activity or alphabetically

Every row shows an all-time count, plus a green +N when there's new activity in the selected timeframe. Click the green number to acknowledge it – it stays acknowledged across restarts until the next real call

Subskills nest under their router with indentation; a router with children doesn't double-count its own calls

Click a skill name to open its description popup – reads the real description: from SKILL.md

Text-triggers and Slash-triggers show as separate lists, parsed straight from the file

Click any slash-trigger to copy it to your clipboard

Add new text- or slash-triggers directly from the popup – writes straight into SKILL.md on disk

Edit mode lets you remove individual triggers one at a time, written back immediately. Remove the last one and the whole Triggers: section is cleaned up, not left empty

Delete Skill archives the whole skill folder (with a confirmation step) into ~/.claude/skills/_archive/skills-retired-<date>/ – never a permanent delete

Pin mode detaches the popup into a real floating panel – stays visible across every Space until you unpin it

Launches at login automatically

Manual refresh, with a last-updated timestamp next to it

⌘Q quits with a real system shortcut