Dashboard
The dashboard is the first screen of onplate. It answers three questions: is everything safe, is everything up to date and how much your skills “weigh” for your AI. The summary tiles lead into the right section with the filter already applied, and the AI system map shows the setup of each tool.
The AI system map
The map brings together the local configurations of Claude, Codex, GLM and OpenCode. For each AI it shows how many resources were found, the approximate permanent weight in context and the number of items that need attention.
The filters above the map let you look at skills, MCP servers, subagents, commands, hooks and rules separately. Click an AI card — a window with the list of resources and the problems found opens over the map. The map itself does not change size, and a long list scrolls inside the window.
From that window you can immediately:
- open the filtered Bindings for the chosen AI and resource type;
- go to the external copies, conflicts or missing files that were found;
- close the window with the cross, the
Esckey or a click on the dimmed area.
The Check again button reads your local AI directories and configs and builds the map. It installs nothing and deletes nothing. After the check, the center of the map shows the time of the report, the total number of resources and how many items need attention.
The top summary
Three headline numbers:
- Tokens always in context — how much “room” your skills permanently take in the AI’s memory (more on that below, in “Context weight”).
- Need security attention — the number of dangerous and suspicious skills. If it is clean you can relax; if there is a number, a click opens the Security scanner.
- Outdated installs — how many installed copies fell behind the version in storage. A click leads to Bindings, where outdated ones are marked “newer available”.
The “Skills” block
Skill counters, each one a link:
| Tile | What it means | Where it leads |
|---|---|---|
| Skills in the catalog | How many skills are in your storage | Skill storage |
| Installed | How many installs were made into AI tools | Bindings |
| Outdated installs | An old version is installed, a newer one is in storage | Bindings filtered to “outdated” |
| External skills | Skills in an AI installed outside onplate | Sync |
| Sync conflicts | The file on disk matches no version | Sync |
| Edited by hand | Somebody edited an installed file directly | Sync |
| Gone from disk | The install is on record, the file is not | Sync |
| Git updates | Connected repositories received new versions | Repositories |
The “Security” block
A summary of the checks: how many skills are safe, suspicious, dangerous and how many are not scanned yet. Two important hints appear here as well:
- “Skills not scanned yet: N — run the scanner” — part of your storage was never analyzed;
- “Skills with outdated acknowledgements: N — the rules got stricter, re-check” — you once marked a finding as safe, but the scanner rules have been updated since, and the decision deserves a fresh look.
If all is well, the block says so: “All clear: nothing dangerous, checks are current”.
The “MCP servers” block
The same counters, but for MCP servers: how many are in the catalog, how many installed, outdated, external and in conflict.
The “Packages” block
How many packages are in storage and how many of them need attention — partly installed, containing dangerous resources or with an update available.
The “Context weight” block
Something only onplate does: an estimate of how many tokens your skills take in the AI context.
How it works: every skill has short metadata (name and description) — that is loaded into every AI session, always. The full body of a skill is pulled in only when the skill fires. The dashboard shows:
- the total “permanent” weight of every globally installed skill;
- the upper bound on activation — the sum of all skill bodies if they all fired at once; this is not a forecast for one ordinary session;
- a list of the heaviest skills — the candidates for a cleanup.
The estimate is approximate (roughly 4 characters per token), but it is enough to notice that the storage has grown fat and it is time to remove the excess.
The “Usage” block
Statistics for the last 30 days: which skills fired and which were never used. It helps you find the ballast — skills that take up context without giving anything back.
Activations are counted in Claude Code, Codex, OpenCode and GLM sessions: activation through the Skill tool and slash commands, reading SKILL.md, $ mentions in Codex, the skill tool in OpenCode. The block heading lists whose logs could be read. The parsing is local only: the logs give the fact — which skill, when and in which project — and not the contents of your conversations.
If the storage is empty
Instead of counters, the dashboard shows a greeting with three starting actions: Import a package, Connect a repository, Scan the system. The Quick start will tell you which to pick.