After data migration (e.g. XDG path change), target symlinks still
point to old source path. PruneOrphanLinks treated these as "external"
and always preserved them, even with --force.
- Auto-remove broken external symlinks (target no longer exists)
- Pass force flag to PruneOrphanLinks; force removes valid external symlinks
- Use utils.PathHasPrefix for Windows case-insensitive path comparison
Use redirect-based version detection (same as install.sh) instead of
calling api.github.com directly, which is rate-limited to 60 req/hr
for unauthenticated requests on shared IPs.
PowerShell 5.1 defaults to TLS 1.0/1.1 which GitHub API rejects,
causing "Failed to get latest version" on Windows. Also surface the
actual error message for easier debugging.
upgradeSkillshareSkill() previously only checked whether SKILL.md
existed, always reporting "Already up to date" regardless of actual
version. Now reads the frontmatter version field and compares against
the remote version, auto-updating when outdated.
All install methods now include the web dashboard via runtime download.
Remove outdated Homebrew-only limitation notes. Add XDG data/state/cache
directories to uninstall instructions.
Move GetCacheDir() from version package to config.CacheDir(), consistent
with BaseDir()/DataDir()/StateDir() pattern including Windows fallback.
Add XDG_CACHE_HOME and cache paths to website docs.
Split ListenAndServe into Listen + Serve so port conflicts are detected
before the browser is opened. Previously, openBrowser ran first and
could connect to an old skillshare ui process on the same port.
- Extract findUnknownSkillTargets() into target_helpers.go, shared by
check and doctor commands
- Reuse buildSkillEntries() + displaySkillsCompact() in project list,
removing duplicate projectSkillEntry type and display functions
- Silence MigrationSkippedDestinationExists warning that fired on every
run after a successful migration
Ubuntu CI runners set XDG_CONFIG_HOME system-wide, causing
config.BaseDir() in subprocesses to resolve outside the sandbox.
Redirect XDG_CONFIG_HOME/DATA_HOME/STATE_HOME to sandbox paths
so every test exercises the XDG codepath with proper isolation.
Separate backups/trash into XDG_DATA_HOME and logs into XDG_STATE_HOME
instead of keeping everything under XDG_CONFIG_HOME. Add automatic
migration at startup to move existing data to the new locations.
Fix Windows path resolution: DataDir() and StateDir() now fall back to
%AppData% on Windows (matching BaseDir()), preventing incorrect use of
~/.local/share and ~/.local/state paths.
Add MigrateWindowsLegacyDir() to move pre-v0.13.0 data from
~\.config\skillshare\ to %AppData%\skillshare\ on Windows.
Update doctor, oplog, backup, trash, and all tests/docs accordingly.
Replace the embedded SPA (go:embed + build tags) with runtime download
from GitHub Releases. The binary no longer bundles UI assets — they are
downloaded on first `skillshare ui` launch and cached at
~/.cache/skillshare/ui/<version>/.
- Delete embed.go / embed_dev.go; add spa_disk.go with placeholder handler
- Simplify ensureUIAvailable(): cache check before dev-mode fallback
- Pre-download UI assets in `skillshare upgrade` (best-effort)
- Remove Node stages from production and CI Dockerfiles
- Keep Node stage in sandbox for playground UI cache pre-population
- Remove ui-embed-check CI job and -tags dev build flag
- Update Makefile, mise.toml, GoReleaser, release workflow
Add config.BaseDir() as single source of truth for base directory
resolution. Replace all hardcoded ~/.config/skillshare paths.
Path resolution priority:
1. SKILLSHARE_CONFIG env var (unchanged)
2. $XDG_CONFIG_HOME/skillshare (any platform)
3. %AppData%/skillshare (Windows only)
4. ~/.config/skillshare (Linux, macOS fallback)
- Add config.BaseDir() with XDG_CONFIG_HOME + Windows %AppData% support
- Update BackupDir(), TrashDir(), init default source to use BaseDir()
- Show resolved base directory in doctor output
- Add unit tests and integration tests for XDG override
- Document XDG_CONFIG_HOME and update Windows paths in docs
When installing from a monorepo where skills live in a nested directory
(e.g. skills/vue-best-practices/), specifying just the skill name as
subdir (owner/repo/vue-best-practices) would fail because the exact
path doesn't exist at the repo root. Now resolveSubdir scans all
SKILL.md files and matches by basename, falling back gracefully with
an ambiguity error when multiple candidates share the same name.
Project targets now use the same short names as global targets (e.g.,
"claude" instead of "claude-code"). Old project_name values are preserved
as aliases so existing configs continue to work.
- Add Aliases field to targetSpec (deprecated, remove once safe)
- Update 11 targets in targets.yaml: project_name = global_name
- LookupProjectTarget falls back to alias lookup
- Fix 3 callers that bypassed LookupProjectTarget
- Extend MatchesTargetName and KnownTargetNames with alias support
- Add TestLookupProjectTarget_Alias
- Update all integration tests, docs, CLI help, and FAQ
- Update Homebrew command from tap (runkids/tap/skillshare) to
homebrew-core (brew install skillshare) in README, website homepage,
and getting-started docs
- Add note that Homebrew build is CLI-only (no web dashboard)
- Add IsUIBuilt() detection in embed.go/embed_dev.go to check for
placeholder UI at startup
- Show CLI warning with install instructions when skillshare ui is
run on a build without the frontend
- Add Homebrew Note section to commands/ui.md with migration steps
Support -p/--project flag with auto-detection, matching other
dual-mode commands. Show include/exclude filters in diff output
so users understand why certain skills are absent from results.
Skip git-not-initialized warning in project mode since the project
itself has version control and backups aren't used. Skip tracked repo
directories (_prefix) in SKILL.md validity check since they are
container directories for nested skills by design.
Each collision detection section now shows both global and project
config filter syntax, with a cross-reference link to the full
target filters documentation.
Collision checks previously ran on all skills before target filters,
producing false positive warnings when users route same-named skills
to different targets via include/exclude. Now collision detection
runs per-target after filtering, downgrading to info when filters
correctly isolate duplicates.
- Add CheckNameCollisionsForTargets() to internal/sync
- Replace collision blocks in sync.go and sync_project.go
- Show human-readable filter summary per merge target
- Add unit tests (5 cases) and integration test
- Update docs with new two-tier collision output examples
Closes#25 — fonts were too fancy and hard to read.
Website (docs):
- Replace Kalam with Inter for all headings and UI elements
- Remove unused Patrick Hand import
- Keep Kalam only for homepage hero title (brand identity)
- Reduce font-weight 700→600 on small uppercase labels
(footer titles, admonition headings, supported CLIs)
Web UI (dashboard):
- Replace Patrick Hand (body) and Kalam (headings) with Inter
- Switch CodeMirror gutter from Patrick Hand to monospace
- Fix include badge contrast: bg-blue-light → bg-info-light
CLI unlinkMergeMode() used exact name matching (sourcePath + entryName)
to identify managed symlinks, which failed for nested skills (flat name
like _team__frontend__ui doesn't match source path _team/frontend/ui)
and orphaned symlinks. Changed to PathHasPrefix-based detection that
checks if a symlink points anywhere under the source directory.
Server handleRemoveTarget() only handled symlink mode (entire dir is a
symlink). Added merge mode branch with unlinkMergeSymlinks() that
iterates target entries and removes symlinks pointing under source.
- Distinguish two dev workflows: make ui-dev (local Go) vs
dev-docker-up + pnpm run dev (no Go required)
- Add auto-init to dev profile so container doesn't crash on first run
- Add --cap-add ALL to playground volume init (cap_drop blocks chmod)
- Document port 19420 conflict between playground and dev profile
- Add docker-build vs docker-build-multiarch explanation and GHCR pull examples
- Increase git command timeout to 180s for constrained Docker networks
- Increase online test timeout to 300s
- Align Makefile help text formatting
Verify that skill identity is derived from directory path (FlatName),
not YAML frontmatter name — ensuring duplicate frontmatter names in
different directories cannot shadow each other during sync.
Hub search was capped at 20 results because it shared the same
normalizeLimit() path as GitHub search. Hub indexes are static JSON
(~100 entries), so returning all results is expected.
- Remove normalizeLimit() from hub search path; limit=0 means no limit
- Backend defaults to limit=0 for hub requests, 20 for GitHub
- CLI uses limit=0 when --hub is set without explicit --limit
- Frontend searchHub() no longer sends a limit parameter
- Replace IIFE rendering with useMemo for stable filter results
- Add IntersectionObserver-based scroll-to-load (20 items per batch)
Hub sources are now saved in config.yaml (both global and project),
replacing the Web UI's localStorage-based approach. This allows CLI
and UI to share the same hub list.
- Add HubConfig/HubEntry types and helper methods (resolve, add, remove)
- Add CLI subcommands: hub add/list/remove/default
- Search --hub now resolves saved labels (e.g. --hub team)
- Add REST API endpoints for hub CRUD (GET/PUT/POST/DELETE /api/hub/saved)
- Migrate SearchPage from localStorage to server API calls
When `skillshare audit <path>` targets an existing file or directory,
config is not needed. Previously config.Load() ran unconditionally,
causing "config not found" errors in CI environments without a
skillshare config. Now path mode bypasses config loading entirely.
Display #tag accent badges on hub search result cards and add an
inline filter input that matches against name, description, and tags.
Filter resets automatically on new searches.
- Fuzzy match on skill name, substring match on description and tags
- Display tags (#tag) in both list mode and interactive selector
- Always show "Search again" option when filtering