* feat(audit): honor findings accepted with --force via --respect-accepted install and update --force record the findings a user accepted in audit_accepted, but audit ignored that record. A skill installed with --force therefore kept failing `skillshare audit` in CI with no way to let the accepted findings pass (#869). audit --respect-accepted marks those findings as acknowledged. They stay in every output (JSON acknowledged: true, a text/TUI note, "(accepted earlier)" in Markdown, SARIF external suppressions), but they are counted in summary.accepted instead of their severity and no longer fail the audit. Without the flag nothing changes, so existing CI gates keep their behavior. A changed file or snippet produces a new key and blocks again. Tracked repositories record acceptances at the repo root with paths relative to it, so ApplyAcceptedFindings walks up from each skill to match them. The install, update and edit gates use the same function and now honor repo-wide acceptances for skills inside tracked repos too. The dashboard always honors accepted findings: the Audit page mutes them and counts them apart, the skill Audit tab collapses them, and the Dashboard attention card no longer lists a skill whose only blocking findings were accepted. The CLI JSON summary and the server summary share Result.CountOpenBySeverity so both count accepted findings the same way. The oplog records the accepted count and the log views show it. * fix(audit): match acceptances when a single file is scanned A single-file scan names its findings by the file's base name, but the ancestor lookup used the file's whole path as the prefix, so it hashed SKILL.md/SKILL.md instead of the recorded SKILL.md. Auditing a skill file directly with --respect-accepted, and auditing agents inside a tracked repository in the CLI or the dashboard, still failed on unchanged accepted findings. For a file target the prefix is now the file's own directory.
32 KiB
sidebar_position
| sidebar_position |
|---|
| 1 |
ui
Launch the web dashboard for visual skill management.
Remote searches (GET /api/search) and install previews (GET /api/preview) use the same origin checks as mutations: cross-site browser requests are refused before fetching. Direct clients without browser origin/fetch headers remain supported. GitHub API credentials are restricted to HTTPS api.github.com and raw.githubusercontent.com on the default port or 443; authorization is removed on cross-origin redirects.
skillshare ui # Run in the foreground
skillshare ui start # Start (or reuse) a background server
skillshare ui stop # Stop the background server
Opens http://127.0.0.1:19420 in your default browser.
Modes
| Mode | Behavior |
|---|---|
skillshare ui (default) |
Runs the UI server in the foreground; Ctrl+C stops it |
skillshare ui start |
Starts the UI server as a background process and returns control to the shell. Re-running start reuses the existing process if it's still healthy |
skillshare ui stop |
Stops the background UI server started by skillshare ui start |
When to Use
- Manage skills, targets, and sync through a visual web interface
- Browse and install skills without memorizing CLI flags
- Run security audits with a visual findings report
- Share a dashboard view with team members less comfortable with CLI
Flags
| Flag | Default | Description |
|---|---|---|
-p, --project |
Run in project mode (uses .skillshare/) |
|
-g, --global |
Run in global mode (uses ~/.config/skillshare/) |
|
--port <port> |
19420 |
HTTP server port |
--host <host> |
127.0.0.1 |
Bind address (use 0.0.0.0 for Docker) |
-b, --base-path <path> |
Sub-path for reverse proxy (e.g., /skillshare) |
|
--trusted-origin <url> |
Another address the dashboard answers at, such as https://tools.example.com behind a reverse proxy. Repeat for more |
|
--no-open |
false |
Don't open browser automatically |
--app |
false |
Open the dashboard as a desktop-style Chromium app window when available (start mode only) |
--clear-cache |
In the foreground form: clear cached UI assets and exit. Combined with start: clear cache, then start in the background |
:::tip Auto-Detection
If .skillshare/config.yaml exists in the current directory, the dashboard automatically starts in project mode. Use -g to force global mode.
:::
Examples
# Default: opens browser on localhost:19420 (foreground)
skillshare ui
# Project mode (manage .skillshare/ skills)
skillshare ui -p
# Custom port
skillshare ui --port 8080
# Docker / remote access
skillshare ui --host 0.0.0.0 --no-open
# Start in the background and return to the shell
skillshare ui start
# Start as a chrome-less desktop-style app window
skillshare ui start --app
# Stop the background server (uses the remembered host/port)
skillshare ui stop
# Clear the cached UI assets, then start fresh in the background
skillshare ui start --clear-cache
Dashboard Pages
The sidebar groups pages by task: syncing, what you manage, where it goes, and upkeep. The line under the name shows the mode and its folder, such as Global · ~/.config/skillshare.
Some pages show a count in the sidebar when they need attention. The counts refresh every 15 seconds while the dashboard tab is open:
- Sync: changes a sync would apply
- Git Sync: uncommitted files, or commits not pushed yet when the tree is clean
- Audit: skills and agents blocked by the last scan (shown after you run one)
| Page | Description |
|---|---|
| Dashboard | Counts for skills, agents, extras, MCP servers, hooks, plugins, and targets, plus items that need attention. Uninstalling a tracked repo with uncommitted changes, or whose git status can't be read, explains why and asks again before moving it to the trash as it is, like uninstall --force |
| Sync | Preview every change per target before writing. Choose which parts to include (Skills, Agents, Extras, MCP). Files edited inside a target are kept unless Force is on. Items that exist only in a target can be collected back to source from here. Each sync backs up target folders first. When a target fails, the others still sync: the page lists each failed target with its part (Skills, Agents, Extras or Config), a plain-language explanation when the cause is a common one (a symlink that points elsewhere, permission denied, a read-only file system, a missing file or folder, invalid target settings) and the error, above the other warnings, offers Turn on Force for a skills symlink that points elsewhere, and marks the target in the change list. Even when every target fails, the page lists them the same way. The Last sync card names the targets that failed in the latest sync. Targets that get the same changes share one card that lists the skills by folder. Discard all moves skills that no target has received yet to the trash, after a confirmation. When a target's include filter matches no skill, a notice says the target gets no skills, and Edit filter opens the place where that filter is set; Git Sync shows the same notice after a pull |
| Git Sync | Commit and push the source repo, push commits that aren't on the remote yet, and pull. Opening the page fetches from the remote, so Pull shows how many commits the remote has. When both computers have new commits, Pull and merge preserves both histories before you push. With uncommitted changes and remote updates, Commit and pull saves your changes first. Pull syncs what the repo scope holds (skills, agents, extras, or root), like pull. Sync both ways commits local changes, pulls and merges, syncs targets, then pushes, like push --pull; a conflict stops it before pushing. Conflicting files open a version comparison; choose the local or remote version of each whole file (or keep all local or all remote at once), then Apply choices and pull. Cancel leaves your files unchanged. When a first pull can't merge with the remote, it offers a force pull that replaces local files with the remote branch |
| Hubs | Reached from the Skills page. Lists the built-in hub, saved hubs, and your own Hubs (Mine); pick one to filter its skills and install them. Add or create a Hub adds an existing hub, creates a new Hub, or imports a skillshare-hub.json. Your own Hubs are changed with Edit; Share downloads the index and builds the hub add command. Refresh reloads a saved or built-in hub's list now instead of waiting for the two-minute cache. See hub |
| Skills / Agents | Installed items, the Updates tab, and the Trash tab. Skills also have an Analyze tab that estimates how many tokens each skill adds to a target's context. Install searches GitHub or installs from a URL or path; its menu also holds New skill, which opens the creation wizard, and Link folder. When the config lists entries that are not installed, the Updates tab and the install dialog offer Install all, like a bare install. After a check, skills deleted from their source repo are listed with Prune and Prune all, which move them to the trash. Link folder links a folder into the skills source like link: it takes a path, an optional link name and, while follow_source_links is off, a checkbox that turns it on, and a refused folder shows the same reason as the command. A linked folder is listed as its own group with the link icon and the folder it points at; the icon button on its row (the link icon turns into an unlink icon on hover) asks for confirmation and then unlinks it like unlink: the folder is not touched and the link can be restored from the Trash tab. A skill inside a linked folder can still be uninstalled on its own, which moves it out of that folder to the trash; linked folders have no Update repo or Uninstall repo. The list and card views can filter by Folder (a tracked repo, a folder such as frontend/react, or Root) and group by Folder, with the root first and folders A→Z. In the list view, a skill inside another skill's folder is listed under that skill: the count button after the package's name opens its skills below it, joined to the package by a faint tree line (packages start collapsed), and deeper packages are indented further. A search opens the packages that hold a match, shows counts such as 1 / 3, and marks the matched text in names; packages you close during a search open again when the search changes, and the packages you opened before stay as they were. Paging counts only top-level rows. Each row shows the description from the skill's frontmatter on one line, in full while the row has keyboard focus; Group and sort → Show → Descriptions hides them. Expand all / Collapse all also opens and closes packages, and the selection bar says how many selected items are not shown. Only a name opens the detail page. Uninstalling a package also moves the skills inside its folder to the trash, and the confirmation says how many. The tree view shows the source folders on the left and details on the right: click a folder or skill to select it, Cmd/Ctrl-click to add to the selection, Shift-click for a range, and double-click a skill to open it. The right side turns everything selected on or off with one switch (an item that a ! rule in .skillignore.local keeps on is reported as failed, with the reason), sets its targets (tracked repos and their subfolders included), and lists each skill with its own switch; a tracked repo also offers Update repo and Uninstall repo. A skill with disable-model-invocation: true carries a manual only tag in the list, on its tile and on its detail page, the same state list toggles with m. A skill installed from a branch other than the remote default (install --branch) shows that branch as a tag in the list, on its tile and on the Updates tab. On the detail page of a skill installed from a remote source (not a tracked repo, which changes branch on Git Sync), Branch lists the source's branches, for any Git remote (HTTPS or SSH, on any host). Picking one asks for confirmation, then reinstalls the skill from that branch with the same security audit and local-changes check as an update. If that branch has no such skill, the install stays as it was. In a project, a successful switch also records the branch in .skillshare/config.yaml, so teammates install from it. In the skill editor, Add field describes what each frontmatter field does. Sync skills / Sync agents previews, then syncs only that kind to every target; the same dialog opens from Sync Now after an update, an uninstall or a collect. The file browser lists SKILL.md and attachments through followed source links, including when the link itself is the skill root. |
| Extras | Folders & files: rules, commands, and other files synced alongside skills. Edit target changes a target's folder, file name, mode, flatten setting and file filters with a preview; Edit extra renames an extra or points it at another source folder. AGENTS.md: in global mode, shared AGENTS.md files and which targets use them; in a project, the project's ./AGENTS.md and whether each target reads it. See Share one AGENTS.md across your tools. Memory: browse, search, preview, edit, and delete shared Markdown notes; add INDEX links, preserve drafts on conflicts, open Backup Files history, and connect agents through reviewed instruction-file changes. Configured status does not prove a read. See the Memory walkthrough |
| MCP | One row per server, with the Agents it syncs to as chips; the count button on the row opens their toggles. Add server takes a URL, a command, a pasted snippet or a file; Import from a target reads what an installed Agent already has. Each server's menu has View what each Agent gets, which lists the source and every Agent's file and shows the one you pick, including unsaved edits. Conflicts offer Import from the Agent or Replace with source. The server dialog's Tools section sets the tool policy: Load tools, which starts the server once with the dialog's settings, saved or not, and lists its tools to tick; an Exclude rules row for * patterns; it lists each selected Agent that does not apply part of the policy, and the row shows a tag when a policy is set. For a Pi server, the dialog also has Pi's built-in MCP settings: tool exposure and other Pi settings, which take piOptions as JSON, with their help in info tooltips. Defaults edits mcp.targets. Sync MCP, in the Sync box, lists the pending changes and writes only the MCP config files, keeping a backup of each; below it, Check checks the servers and Backups and restore previews and restores those backups. |
| Plugins | One row per plugin, with its Agents as toggles. Expanding a row also lists the other Agents the source supports; ticking one previews an install. The row menu syncs, updates, removes, or opens View files, a read-only browser of the local copy Skillshare reviewed. Share copies one command that adds the ticked plugins from their HTTPS sources. See Manage plugins across tools |
| Targets | Target list with status. Add target also takes Another account: a second config folder of an Agent you already use, with a preview of where it writes. Each target's page edits include/exclude filters and collects local-only skills back to source. A search box above the filter preview narrows its rows by name and marks the matching text; it changes only what is shown, not the filters. Project pages and the Extras Edit target dialog share this preview and its search. The Syncing column shows what each target gets, as links: Skills, Agents, MCP and Hooks with their counts, Extensions for Pi, and the target's instruction file when a shared AGENTS.md is connected to it; each opens that tab of the target's page. A target that has an MCP config file gets an MCP tab with one row per server; a click saves at once, and Sync all targets writes the MCP files of every target. Each target also has a tab named after its instruction file (CLAUDE.md, GEMINI.md, AGENTS.md, ...) that shows the read order, edits the file and converts it to AGENTS.md |
| Projects | Global mode only. Project folders that the global config syncs into, from projects in config.yaml, projects kept on this machine only (marked This machine), and mcp.projects. The same projects can be managed from the terminal with skillshare project. Add project takes a folder, its targets and what to sync, and saves it in config.yaml unless you pick This machine only under Saved to at the bottom of the dialog (the default when projects_storage is machine-local). Edits are saved where the project is. To move a saved project, use skillshare project localize or share. Removing a project that this machine overrides asks whether to remove only this machine's copy or config.yaml's as well. Each project has Skills and Agents tabs with filters, a preview and the folders that get written, and an MCP tab that turns global servers off in that folder or gives it servers of its own, with Check for those servers. Sync project previews, then syncs only that project's skills, agents and MCP. Targets that already point into a project folder can be converted |
| Audit | Security scan of skills and agents, with findings by severity. When a followed source link could not be read, the counts are marked Scan incomplete beside the Skills and Agents tabs, so a partial scan is not mistaken for a full one; the scan then warns instead of reporting all clear, and the Dashboard lists it under Needs attention. The Rules tab browses every rule by category: switch one off, change its severity, apply a severity to a whole category, pick the scan profile (default, strict, permissive), or open the editor for custom audit-rules.yaml |
| Settings | Tabbed: General (source paths, sync mode, appearance), Backup (target folder snapshots, earlier versions of files such as AGENTS.md, and MCP config backups; see backup), Log (operation history), Health (the same checks as doctor), Extensions (sync-time file transforms), Files (direct editors for config.yaml, .skillignore, and .agentignore; when a .skillignore.local or .agentignore.local exists, the editor shows its rules above the file, since they apply after it and can override it) |
Discard changes beside the change list restores all tracked files and the index in the selected Git scope to the last commit and deletes untracked files and folders, after confirmation. Ignored files, nested Git repositories, and root-scope config.yaml are kept. It does not change commits or push; the action cannot be undone. Dry run previews without changing files. A first commit is required.
The Git conflict comparison uses complete file versions. Choosing a deleted version deletes that file in the merge; non-conflicting files merge normally. Both commit histories remain available. Metadata conflicts resolve automatically. Binary files and files larger than 16 KiB can be selected but have no text preview. If either reviewed revision changes before you apply your choices, the dialog refreshes and requires new choices. After the merge, push to share the result with your other computer.
On the Updates tab, a progress bar tracks the update run and the active row is marked while it updates. Blocked or failed updates appear in a separate section. After a successful update, the tab checks for pending sync changes of that kind. Sync Now appears only when there are changes to apply; otherwise it reports that no sync is needed. Existing symlinks already read updated source content. If the sync check fails, the tab shows the error and offers Retry instead of reporting that everything is up to date. Followed Git checkouts are listed as informational rows with their link name and target path. Like skillshare update --all, this tab does not check, update, or force-retry them; manage those checkouts yourself or use an explicit CLI update by name.
Old links such as /collect, /install, /search, /trash, /analyze, /backup, /log, and /doctor redirect to their new place.
The Files tab puts a panel beside the editor. For config.yaml it shows what the field under the cursor does, the file's structure, and the unsaved changes; for the ignore files it lists what the patterns currently hide. Cmd+S / Ctrl+S saves. The rules editor under Audit -> Rules -> Edit YAML has the same panel plus a Test tab that runs a rule's regex against lines you paste.
Theme System
The dashboard supports two visual styles and three color modes, switchable via the Theme button in the sidebar:
| Setting | Options | Default |
|---|---|---|
| Style | Clean (professional), Playful (bold outlines, hard shadows, handwritten headings) |
Playful |
| Mode | Light, Dark, System (follows OS preference) |
Light |
Theme preferences persist in localStorage across sessions.
Project Mode Differences
When running in project mode (-p), the dashboard adapts:
- Sidebar shows
Project · <project path>under the name - Git Sync page is hidden (project skills use the project's own git)
- Sync backs up agent target folders only, like
skillshare sync -p - Backup tab is hidden in Settings (use version control instead)
- Tracked Repos section is hidden from Dashboard (not applicable)
- Settings -> Files shows
.skillshare/config.yamland the project-level.skillignoreinstead of the global versions - Available targets lists project-level targets (e.g.,
.claude/skills/relative to project root) - Targets count and switch the project's MCP servers, written to the project's own files (such as
.mcp.json). For Claude Code, OpenCode, Kilo Code and Pi, the MCP tab also lists the switches that turn a global server off in this project - Install automatically reconciles
skills:entries in the project config - Extras -> AGENTS.md edits the project's
./AGENTS.mdinstead of shared files, and offers a small fix for targets that only read their own file
UI Preview
REST API
The web dashboard exposes a REST API at /api/. All endpoints return JSON.
| Method | Path | Description |
|---|---|---|
| GET | /api/overview |
Skill/target counts, mode, version, config folder (configDir) |
| GET | /api/skills |
List all skills with metadata |
| GET | /api/skills/{name} |
Skill detail + SKILL.md content |
| GET | /api/skills/templates |
Get available patterns and categories for skill creation |
| POST | /api/skills |
Create a new skill (name, pattern, category, scaffoldDirs) |
| DELETE | /api/skills/{name} |
Uninstall a skill |
| GET | /api/targets |
List targets with status, include/exclude filters, and per-target expected counts |
| POST | /api/targets |
Add a target |
| DELETE | /api/targets/{name} |
Remove a target |
| POST | /api/sync |
Run sync (supports dryRun, force, kind, and project, a declared project root that limits the sync to that project's targets). Backs up targets first unless dryRun is set |
| POST | /api/git/commit |
Create a local git commit from the source repo without pushing |
| POST | /api/git/discard |
Discard uncommitted changes in the configured Git scope (global mode only; requires a first commit). Supports dryRun. Keeps ignored files, nested Git repositories, and root-scope config.yaml |
| GET | /api/git/status |
Source repo status, including commits not pushed yet (ahead) and upstream commits not pulled yet as of the last fetch (behind). Never fetches |
| POST | /api/push |
Commit any changes, then push. Sets the upstream on the first push. When the remote has commits this repo lacks, it fails with 409 and error code push_rejected; pull, then push again |
| POST | /api/pull |
Pull, then sync what the repo scope holds. Diverged history is merged; .metadata.json conflicts resolve automatically. Other conflicts return 409 pull_conflict with the merge undone and error_params containing localHash, remoteHash, and files with local/remote previews. Retry with resolution: { localHash, remoteHash, choices }, where choices maps every conflicting path to local or remote. Stale or incomplete choices return a fresh conflict review. Supports dryRun. When a first pull can't merge, it fails with error code merge_failed; retry with force: true to replace local files with the remote branch. alwaysSync: true syncs targets even when nothing new was pulled. A remote with no branches returns 400 remote_empty |
| GET | /api/diff |
Diff between source and targets |
| GET | /api/search?q= |
Search GitHub for skills |
| POST | /api/install |
Install a skill from source |
| GET | /api/audit |
Scan all skills for security threats; adds warnings and incomplete: true when a followed source link could not be read, as audit --format json does. Findings accepted earlier with --force carry acknowledged: true and are counted in summary.accepted instead of their severity |
| GET | /api/audit/rules |
Get custom audit rules YAML |
| PUT | /api/audit/rules |
Save custom audit rules (validates regex) |
| POST | /api/audit/rules |
Create starter audit-rules.yaml |
| GET | /api/audit/rules/compiled |
Every rule after merging built-ins with custom rules, plus the active profile |
| POST | /api/audit/rules/toggle |
Enable, disable, or re-rate a rule or a whole pattern |
| POST | /api/audit/rules/reset |
Delete custom rules and restore the built-in defaults |
| PATCH | /api/audit/policy |
Set blockThreshold, profile, or both |
| GET | /api/log |
List log entries with optional filters |
| GET | /api/config |
Get config as YAML |
| PUT | /api/config |
Update config YAML |
| GET | /api/skillignore |
Get .skillignore content, .skillignore.local content when present, and ignore stats |
| PUT | /api/skillignore |
Update .skillignore content |
| GET | /api/doctor |
Run all health checks (JSON) |
| GET | /api/doctor/stream |
Run all health checks as Server-Sent Events: a check event for each check as it is found, then done with the same JSON as /api/doctor |
| GET | /api/health |
Liveness probe; returns 200 once the server is ready |
| GET | /api/version |
Current/latest version, whether an upgrade is available, and, for a Nix install, the command that updates the CLI (nixUpgradeCommand) |
| POST | /api/upgrade |
Run skillshare upgrade in place (returns devMode: true when the binary is a dev build, and 409 with code upgrade.nix when a Nix install has a newer CLI waiting) |
| POST | /api/restart |
Restart the local UI server; optional { "clearCache": true } body clears cached UI assets first |
In-place Upgrade
When the dashboard detects that a newer CLI release is available, the Update dialog and the Doctor page's Version card both expose an Update now button:
- The UI calls
POST /api/upgrade, which runsskillshare upgradeon the host. - Once the new binary is in place, the UI calls
POST /api/restartto restart the local server. - The browser polls
GET /api/healthand reloads automatically once the new server is ready.
If skillshare is installed with Nix, the endpoint still updates the built-in skillshare skill but leaves the CLI to Nix. When a newer CLI is available, it answers 409 (upgrade.nix) and the dialog then shows how to update through Nix. Before you click, the dialog already notes how Nix updates the CLI: nix profile upgrade skillshare for a nix profile install, or a rebuild of your Nix configuration. See upgrade.
If the running binary is a development build (version == "dev"), the upgrade endpoint returns devMode: true and the UI simulates a restart without modifying anything on disk.
If the auto-reload doesn't complete, the dialog tells you to run skillshare ui start to bring the background server back up.
Reverse Proxy
If you run the dashboard on a shared server behind a reverse proxy (e.g., homelab, internal tools platform), pass the address you open in the browser with --trusted-origin, and use --base-path to serve it under a sub-path alongside other services:
skillshare ui --base-path /skillshare --host 0.0.0.0 --trusted-origin https://tools.example.com --no-open
Or via environment variables:
SKILLSHARE_UI_BASE_PATH=/skillshare SKILLSHARE_UI_TRUSTED_ORIGINS=https://tools.example.com skillshare ui --host 0.0.0.0 --no-open
The dashboard answers only at localhost, an IP address such as http://192.168.1.20:19420, the name given with --host, and each trusted origin. Any other domain name gets 403, because a DNS rebinding attack points a domain name it controls at your machine. Give each origin as a scheme and host, plus the port when it is not the default, such as https://tools.example.com or http://nas.local:8080. Repeat --trusted-origin, or separate origins with commas in SKILLSHARE_UI_TRUSTED_ORIGINS. Anyone who can reach a trusted address can use the whole dashboard, including the MCP, Hooks and Plugins pages that write commands Agents run, so list only addresses you control and restrict who can reach them at the proxy.
Nginx
location /skillshare/ {
proxy_pass http://127.0.0.1:19420;
proxy_set_header Host $http_host;
proxy_set_header X-Real-IP $remote_addr;
}
Caddy
handle_path /skillshare/* {
reverse_proxy 127.0.0.1:19420
}
Keep the browser's Host header, including its port, as both examples do. The dashboard matches Host against the trusted origins, and refuses a change when the request's Origin does not match its Host, so another website cannot make changes through your browser. Caddy keeps Host by default; do not rewrite it with header_up Host.
:::tip
Without --base-path, the dashboard behaves identically to before — no configuration needed for direct access on localhost:19420.
:::
Docker Usage
To use the web UI inside Docker (requires network access for first-time UI download):
make playground
# Inside container:
skillshare ui --host 0.0.0.0 --no-open
Then open http://localhost:19420 on your host machine (port 19420 is mapped automatically).
Project Mode
The web dashboard fully supports project-level skills:
cd my-project
skillshare ui -p
Or simply skillshare ui if .skillshare/config.yaml exists (auto-detected).
The dashboard reads and writes .skillshare/config.yaml, syncs to project-local targets, and reconciles remote skill entries after install — just like the CLI.
Runtime UI Download
skillshare ui automatically downloads pre-built UI assets from the matching GitHub Release on first launch. The assets are cached in ~/.cache/skillshare/ui/<version>/ (respects XDG_CACHE_HOME) so subsequent launches are instant and offline.
- First run requires an internet connection to download the UI assets (about 2 MB)
- Subsequent runs use the cached assets — no network needed
- On upgrade,
skillshare upgradepre-downloads the new UI and keeps both the new and pre-upgrade versions cached; other cached versions are cleaned up - To clear the cache manually, run
skillshare ui --clear-cache
Homebrew Note
All install methods (Homebrew, Nix, installer script, manual binary) use runtime UI download. When you run skillshare ui, it automatically downloads the UI assets from GitHub on first launch. After that, the cached assets are used offline.
To clear the downloaded UI cache:
skillshare ui --clear-cache
Architecture
The web UI is a single-page React application downloaded at runtime from the matching GitHub Release and served from disk cache (~/.cache/skillshare/ui/<version>/).
skillshare ui
├── Go HTTP server (net/http)
│ ├── /api/* → REST API handlers
│ └── /* → Cached React SPA (runtime download)
└── Browser opens http://127.0.0.1:19420
See Also
- status — CLI status check
- sync — CLI sync command
- Project Setup — Project mode setup guide
- Docker Sandbox — Run UI in Docker






