Separate user-owned followed Git checkouts into linked_repos in check and resources responses so the Updates tab renders information instead of tracked errors or retry actions. Refuse dashboard updates before Git runs, including force retries and descendant skills, while leaving managed repositories unchanged.
Users link a checkout directly under the skills source (e.g.
skills/_dev-skills -> ~/code/dev-skills) and expect it to behave like any
other first-level entry. sourcewalk resolved only the root, so these links
were invisible to discovery, sync, update, and the dashboard.
follow_source_links (global and project config, default false) enables a
per-operation sourcewalk.Follow policy. Walk, WalkDir, and ReadDir report a
followed link under its own name as a directory and descend into its
target, so names, RelPath, FlatName, SourcePath, tracked-repo detection, and
metadata keys stay logical. Only entries directly under the source root are
followed. The policy is built from config and passed explicitly to every
discovery caller in the CLI, server, and reconcile, so counts agree; the
server keeps a per-instance snapshot refreshed on config load and save
instead of a process-global flag.
A link is skipped, with one warning per operation, when its target is
missing or unreadable, is the source or a parent of it, or overlaps an
active skills target. A missing or unreadable target marks the operation
incomplete: sync then skips pruning and copy-orphan deletion, and reconcile
keeps metadata, so an unmounted drive or a re-clone in progress cannot
erase target entries or install records. A link to an existing file is not
a directory link and stays an ordinary entry; treating it as incomplete
would disable pruning for anyone sharing a file such as .skillignore.
doctor keeps raw link identity: with the key off it still says the link is
not followed and names follow_source_links: true; with it on it reports
followed links and skipped links with the reason.
The raw-walk allowlist gains entries for the policy path in Walk and
WalkDir: the source root walk with a policy, and the walk of a validated
link target whose paths are remapped to the logical link path. Both live
in sourcewalk itself, the package that owns source traversal.
Refs #274
* fix(check): compare project-relative local sources and record install layout
Project installs record a local source as typed (./vendor/skill), relative
to the project root, so check skipped them. Resolve them against the project
root in both the CLI and the dashboard.
The copied shape of a local install was inferred from its file hashes, so a
CLI install whose source held only SKILL.md looked like a dashboard
collection-root install: a child skill added at the source later was neither
reported by check nor copied by update. Record the layout (skill-file or
directory) at install time and keep the hash heuristic only for entries
written before the field existed.
* fix(install): let a recorded layout decide the update shape without hashes
An entry whose hashing failed at install time still knows what was copied;
returning false there would turn a SKILL.md-only update into a full copy.
* feat(check): detect changes at local install sources
Skills installed from a local path (for example a skill shipped inside an
app bundle) could already be refreshed with `update`, but `check` always
reported them as "local source", so users had no signal that the source
had changed after an app update.
`check` now re-hashes the files at the recorded source path and compares
them with the hashes stored at install/update time. It reports
update_available on a difference, up_to_date on a match, and an error
naming the path when the source is gone. The path is resolved the same
way `update` resolves it before re-copying, so a successful update always
brings the status back to up_to_date. The CLI and both Web API check
handlers share the comparison in internal/check.
Refs #206
* fix(check): avoid false local-source updates and show check errors
Review follow-ups for local-source change detection:
- The dashboard installs the root of a directory that also holds child
skills as its SKILL.md alone, so hashing the whole source reported
update_available right after a clean install. Compare only the files
such an install copies, recognized by recorded hashes that contain none
of the child skills present in the source.
- A relative source path resolves against the cwd of the install, which
is not recorded, so checking from another directory reported a missing
source. Relative sources are no longer compared.
- The dashboard dropped the new check `message`, so a missing source only
showed a generic "Check failed". Carry it into the status tooltip.
* revert: restore video files accidentally removed in 1fb13b75
The review-fix commit picked up unrelated staged deletions under video/. They belong to separate work and are not part of this change.
The dashboard check fetched the remote HEAD without credentials when a
skill had no pinned branch. Private repos therefore reported an error in
the dashboard while the CLI check, which uses
GetRemoteHeadHashWithAuth, succeeded. The pinned-branch path already
used auth, and the default-branch path now matches it.
The dashboard check reimplemented the CLI's per-repo check but fetched
without the HTTPS token and ignored git status errors. Private tracked
repos could fail in the dashboard while `skillshare check` worked, and a
repo whose status could not be read showed as up to date. Both check
endpoints now call check.ParallelCheckRepos, the same code the CLI uses.
The JSON and SSE response shape does not change.
Batch uninstall treated a failed dirty check as clean and moved the
repo to trash. Without force, that item now fails with the git error
instead. This is stricter than the CLI, which only warns and continues.
Batch target updates ignored metadata save failures and still reported
success, so an override could silently disappear on the next reload.
They now return 500.
Every API request reloads the skills and agents metadata stores and
swaps the server's pointers under s.mu, but several handlers read or
wrote the stores outside that lock: listing, skill detail, check, the
update stream, and the hash refresh after setting targets. Concurrent
requests could race on the pointer or on the stores' maps, which the
race detector reported hundreds of times under load and which can
crash the server with a concurrent map access.
Store access now always happens under s.mu. Reads return a copy of
the entry so it can be used after unlocking, and target handlers hash
files outside the lock and only write the result under it.
Skills installed from a non-default branch record that branch in metadata,
but the grouped update path (update --all) cloned the remote default branch,
so the skills were either downgraded or reported as removed upstream and
pruned. The web UI update and check handlers had the same gap, comparing
against the remote HEAD instead of the installed branch.
Group batch updates and web UI checks by URL+branch, pass the branch through
to the clone, and resolve remote tree hashes for that branch as well.
Refs #268
Add ProjectSources struct to ProjectConfig with optional skills, agents,
and extras fields. When set, project-mode commands resolve source paths
from these values instead of hardcoded .skillshare/{skills,agents,extras}/.
Config example:
sources:
skills: ./docs/skills
agents: ./docs/agents
Three EffectiveXxxSource(projectRoot) methods provide the central
resolution logic with fallback to defaults. The projectRuntime gateway
and server layer route through these methods, so most commands get
custom source support automatically.
Closes#153
All call sites that looked up metadata entries using filepath.Base()
now use store.GetByPath(relPath) which handles both full-path keys
and legacy basename+group keys transparently.
Also fixes makeInstallOpts() to set SourceDir from updateContext,
preventing nested skill updates from writing metadata to wrong dir.
Replace every remaining call to the old per-skill metadata API
(ReadMeta, RefreshMetaHashes, config.LoadRegistry) with the centralized
MetadataStore across 10 files:
Server handlers:
- handler_check.go: use s.skillsStore.Get() instead of ReadMeta
- handler_check_stream.go: same pattern as handler_check.go
- handler_skills_batch.go: use s.skillsStore.RefreshHashes() + Save()
CLI commands:
- check.go: load store once in collectCheckItems, use MetadataEntry
- doctor.go: load store once in checkSkillIntegrity
- list.go: load store once in buildSkillEntries (replaces parallel ReadMeta)
- uninstall.go: use skillsStore for display/dry-run/finalize
- uninstall_project.go: same pattern, remove config.LoadRegistry
- update_batch.go: pruneRegistry now uses MetadataStore
Hub:
- hub/index.go: load store once in BuildIndex
The only remaining ReadMeta call is in a test file
(search_batch_meta_roundtrip_test.go) which tests the shim itself.
handler_sync: add Kind field to diff stream item struct.
handler_audit: add Kind field to audit stream result.
handler_check: add Kind field to check stream result.
handler_update: add Kind field to update stream result.
handler_install: add Kind field to install request for agent support.
Addresses P1 and P2 from code review on the RWMutex commit:
P1 — Stream handlers now snapshot config under RLock before streaming:
- handleDiffStream: clone Targets map to prevent concurrent map panic
- handleCheckStream, handleAuditStream: snapshot source/projectRoot
- handleUpdateStream: snapshot source (per-item Lock already correct)
P2 — All read handlers now use narrow snapshot instead of defer RUnlock:
- RLock → copy needed fields to locals → RUnlock → do work with locals
- Prevents slow I/O (git fetch, version check, audit scan) from blocking
the auto-reload middleware's exclusive Lock on concurrent requests
Other changes:
- computeTargetDiff takes source parameter (no more s.cfg.Source access)
- Add cloneTargets() helper for safe map copying
- Add writeOpsLogTo/writeAuditLogTo for lock-free oplog writes
The // subdir syntax (e.g., repo//skills/foo) stores meta.Subdir with
a leading "/". git rev-parse and ls-tree paths never have leading
slashes, causing tree hash lookups to fail silently.
Strip leading "/" in getSubdirTreeHash(), resolveSkillStatuses(), and
handler_check.go. Add LeadingSlash unit test and E2E runbook.
Previously `skillshare check` compared repo-level commit hashes, causing
all skills from a registry repo to show "update_available" whenever any
commit was pushed — even if the individual skill's files were unchanged.
Now each skill records its git tree hash (subdir-level SHA) at install
time. During check, a three-tier comparison is used:
1. Fast path: commit hash matches → up_to_date (zero extra cost)
2. Tree hash path: blobless fetch (~150KB) + batch ls-tree → per-skill
comparison; unchanged subdirs stay up_to_date
3. Fallback: no tree hash (old installs) → commit comparison (unchanged)
Also fixes pterm progress bar printing twice when it auto-stops at 100%.
- Add (project) labels to status, sync, collect, target, and uninstall
headers for clear mode distinction (matching list/target list pattern)
- Add ui.WithModeLabel() helper for shared functions that run in both modes
- Update uninstall confirmation prompt: "from the project?" in project mode
- Add "Privacy-first" to README Why section and website homepage
- Replace "Unified Source" card with "Privacy-First" card on website
- Unify Quick Actions card heights with h-full in DashboardPage
- Add frontend rebuild note to Docker sandbox docs