Production code no longer reads or writes .skillshare-meta.json sidecars:
- InstallAgentFromDiscovery: write to centralized store via WriteMetaToStore
- agent_check.go: read from centralized store passed as parameter
- handler_skills.go: read agents from s.agentsStore
- audit.go: readMetaFileHashes reads from parent .metadata.json
- analyzer_metadata.go: extract findMetaEntry to deduplicate walk logic
- uninstall_agents.go: remove entry from centralized store on uninstall
- batchUpdateAgents: pass store to reinstallAgent, remove dead field
- Replace LoadMetadata+nil guard with LoadMetadataOrNew across 6 files
- Remove MetadataEntry.Name (json:"-", never set in production) and
FullName() method (only used in one test)
- Add LoadMetadataOrNew() helper to replace 8+ instances of the
LoadMetadata + nil guard boilerplate
Metadata store entries for grouped skills (e.g. frontend/foo, backend/foo)
were keyed by basename only, causing the second install to overwrite the
first. This changes the key to the full relative path.
Key changes:
- WriteMetaToStore: use full rel path as key, clean up legacy basename key
- Add KeyToRelPath() helper and MigrateLegacyKey() method for backward compat
- Reconcile: migrate legacy basename keys to full-path keys on load
- install_queries: guard against double-prefixing with KeyToRelPath
- LoadMetadata: auto-run sidecar cleanup on every load (idempotent)
- migrateAgentSidecars: recurse into subdirectories
- mergeSkillSidecar: use full-path keys, detect tracked repos
Code quality fixes from review:
- Extract RemoveByNames() to eliminate copy-paste in uninstall Phase 7
- Add MetadataKindSkill/MetadataKindAgent constants
- Remove dual-write shim (findSkillsRoot + writeMetaToCentralized)
- install_apply.go uses WriteMetaToStore with SourceDir (no sidecar)
- Move LoadMetadata out of walk loops (P0 efficiency fix)
- isGroupDir accepts store parameter to avoid redundant loads
- Add SourceDir to InstallOptions for --into group support
- Reconcile tests use bare name as store key (not group/name)
- updatable_test uses MetadataStore directly instead of WriteMeta dual-write
- Tracked repos marked with Tracked: true in store
- All unit + integration tests pass (make check green)
- Dual-write in WriteMeta: also writes to .metadata.json for backward compat
- Migration looks in parent dir for registry.yaml
- Group uninstall matches entries by Group field + full path
- Sync test uses MetadataStore instead of registry.yaml
- All integration tests pass (0 FAIL)
ComputeFileHashes now skips both .skillshare-meta.json and
.metadata.json. WriteMeta, ReadMeta, HasMeta, NewMetaFromSource,
RefreshMetaHashes marked as deprecated.
LoadMetadataWithMigration auto-detects old format on first load:
reads registry.yaml entries, merges sidecar .skillshare-meta.json
fields, writes unified .metadata.json, cleans up old files.
Idempotent — skips migration if .metadata.json already exists.
Introduces MetadataEntry (merges SkillMeta + RegistryEntry) and
MetadataStore (map-based container with Set/Get/Remove/Has/List).
This is the foundation for centralized .metadata.json files.
Backend (install_discovery.go):
- Add discoverAgents() call to all 3 subdir discovery paths (sparse
checkout, GitHub API, full clone fallback). Previously only whole-repo
discovery detected agents.
Frontend (InstallForm.tsx):
- Single-skill fast path now checks hasAgents before auto-installing.
When agents exist alongside a single skill, falls through to the
kind selector instead of silently installing as skill.
installTrackedRepoImpl: discover agents alongside skills after clone.
Report agent count in warnings. Update empty-repo warning to account
for repos that have only agents.
updateTrackedRepo: discover agents after git pull, report count.
- Enable frontmatter parsing in DiscoverSourceSkillsAll so /api/skills
returns saved targets (fixes target display reverting to All)
- Skip tracked-repo skills in batch/single target handlers to prevent
dirty git state; reject with 400 for single-skill endpoint
- Add RefreshMetaHashes to recompute audit hashes after SKILL.md edits;
batch handler refreshes outside the write lock
- Show 'Uninstall Repo' with confirm dialog for tracked-repo members
instead of the always-failing individual uninstall action
- Return 409 Conflict when enabling a skill disabled by glob pattern
in .skillignore/.skillignore.local instead of false success
- Preserve IsInRepo flag for root-skillignore-disabled skills so they
remain visible in the Tracked tab after disable
- Update changelog and website changelog for v0.18.7
P1: check.go '@' separator broke SSH URLs (git@github.com:org/repo)
— splitURLBranch would split at the '@' in the URL itself
— Fix: use tab character as separator (cannot appear in URLs)
— Added unit test for SSH URL round-trip
P1: install_config.go grouped same-repo different-branch skills together
— groupConfigSkillsByRepo keyed by CloneURL only, ignoring Branch
— Fix: include branch in grouping key so each branch clones separately
P2: reconcile never cleared stale branch from registry
— Only updated when meta.Branch was non-empty, never cleared it
— Fix: sync branch unconditionally when meta is available
sparseCloneSubdir in discoverFromGitSubdir was hardcoding empty branch,
ignoring --branch when sparse checkout was used for subdirectory discovery.
Reported by: gemini-code-assist on PR #112
- Add GetRemoteRefHash/WithAuth/WithEnv to internal/git/info.go:
uses git ls-remote with refs/heads/<branch> when branch is set,
falls back to HEAD when branch is empty
- Add Branch field to URLCheckInput and URLCheckOutput in check/parallel.go;
ParallelCheckURLs dispatches to GetRemoteRefHashWithAuth when Branch is set
- check.go: group skills by RepoURL@branch composite key so skills on
different branches of the same repo are checked against the correct ref;
urlHashMap keyed by same composite key for correct broadcast
- install_config.go (InstallFromConfig): propagate SkillEntryDTO.Branch
into Source.Branch for both plain and tracked skill phases so registry
rebuild (install --from-config) respects the recorded branch
- NewMetaFromSource: set meta.Branch from source.Branch
- installFromDiscoveryImpl: forward discovery.Source.Branch when building
per-skill Source struct for metadata
- installFromGitSubdir: pass source.Branch to sparseCloneSubdir (was
hardcoded empty string, causing subdir installs to ignore --branch)
- ReconcileGlobalSkills: sync meta.Branch to registry on both new and
existing entries
- toSkillEntryDTOs: include Branch field in DTO conversion
- updateRegularSkill / updateSkillFromMeta: restore source.Branch from
meta before reinstalling, so updates pull from the originally-specified
branch
Tests: TestInstallBranch_MetadataPersistence, TestInstallBranch_UpdatePreservesBranch
Source.Branch threads the branch parameter through discovery.
cloneRepo() passes --branch to git clone when set.
source.Branch is set from parsed.opts.Branch in cmdInstall.
Ref: #90
cloneTrackedRepo(), cloneRepoFull(), and sparseCloneSubdir() now accept
a branch parameter. When non-empty, --branch <name> is passed to
git clone. Fallback chain (partial → full) preserved.
Ref: #90
Replace the naive string matcher (~40 lines) with a proper gitignore
implementation (~200 lines, no external deps) supporting **, ?, [abc],
!negation, /anchored, pattern/ (dir-only), and \#/\! escapes.
New Matcher API (Compile, ReadMatcher, Match, CanSkipDir, HasRules)
replaces the old Match/ReadPatterns functions (kept as deprecated
wrappers). Call sites in sync.go and install_discovery.go updated.
Also fixes isDir=true for install discovery (skill paths are dirs).
100% test coverage on skillignore package (~150 table-driven cases).
Added 7 sync discovery tests and 5 install integration tests covering
**, !negation, pattern/, [Tt] char class, and combined patterns.
DiscoverSourceSkills and DiscoverSourceSkillsLite now load .skillignore
from tracked repos and skip matching paths during walk. This prevents
false warnings in doctor/status/list for vendored SKILL.md files
(e.g. .venv/ inside a tracked repo).
Extract shared skillignore package from internal/install so both
install-time and source discovery use the same filtering logic.
Enable teams to use one project as a shared skills repo where each
developer manages their own targets independently.
- Add `--config local` flag to `init -p` that gitignores config.yaml
- Auto-detect shared repo on clone via .gitignore check (no flag needed)
- Fix partialInitRepair to run interactive prompt instead of auto-select
- Export GitignoreContains, add UpdateGitIgnoreFiles for file-mode entries
- Refactor gitignore helpers to share addGitIgnoreEntries private helper
- Add 5 integration tests covering all new flows
- Add recipe docs page, update init reference, changelogs, release notes
The update handler used internal/git which ran git commands without
capturing stderr, so failures like divergent branches only showed
"exit status 128". Now runGitWithProgress captures stderr and pipes
it through install.WrapGitError for meaningful messages.
Also adds:
- extractGitFatal() strips verbose hint: lines from git output
- Force checkbox and per-item Force retry button in Update UI
- "(try force update)" / "(try --force)" hint on pull failures
When git commands fail with an empty stderr (common in --progress mode),
wrapGitError() was returning the raw exec.ExitError which only produces
"exit status 128" — giving users no clue about the actual problem.
Now extracts the exit code and maps it to actionable messages:
- exit 128: "repository not found or authentication required"
- exit 128 + token: "authentication token was rejected"
- exit 1: "check network connectivity and repository URL"
Repos like pbakaus/impeccable commit .claude/skills/ alongside their
source skills, causing discoverSkills to find every skill twice (36
instead of 18). Fix by deriving known target dotdir names (.claude,
.cursor, .skillshare, etc.) from targets.yaml project_path values and
skipping them during filepath.Walk.
Non-target hidden dirs (.curated, .system — used by openai/skills) are
still discovered as before. Injection pattern avoids circular import
between config and install packages.
JihuLab (jihulab.com) is a GitLab distribution that supports nested
subgroups but its hostname doesn't contain "gitlab". Add "jihulab" to
the built-in host detection so URLs like jihulab.com/group/sub/project
are parsed as full-path repos without requiring gitlab_hosts config.
Support CI/CD pipelines that need gitlab_hosts without a config file.
The env var is comma-separated and merged (deduplicated) with config
file values in both global and project modes.
Also cleans up from review:
- Extract isValidGitLabHostname to share validation logic
- Remove redundant ToLower before EqualFold in isGitLabHost
- Revert 3 no-opts call sites back to ParseSource()
- Add parseOptsFromProjectConfig helper to reduce inline duplication
- Fix project mode LoadProject missing env var merge
- Add install.md tip about custom GitLab domains
Self-managed GitLab instances on custom domains (e.g., git.company.com)
without "gitlab" in the hostname were parsed with 2-segment owner/repo
split instead of full-path repo. Users can now declare extra GitLab
hostnames in config so the parser treats them like gitlab.com (nested
subgroup support).
- Add `gitlab_hosts` field to both global and project Config structs
- Validate entries at load time (reject scheme, path, port, empty)
- Add `ParseOptions` / `ParseSourceWithOptions` to thread hosts into parser
- Add `isGitLabHost()` helper combining built-in + custom host detection
- Extend `InstallContext` interface with `GitLabHosts()` method
- Update all ParseSource call sites (install, search, update, server)
- Add JSON schemas for both config files
- Add unit tests for config validation and source parsing
- Add E2E runbook for devcontainer verification
- Document in configuration reference and URL formats appendix
The initial GitLab subgroup fix treated all multi-segment HTTPS paths as
full repo paths, regressing GHE and generic host subdir installs like
github.mycompany.com/org/repo/path/to/skill.
Add host-based heuristic: hosts containing "gitlab" use full-path-as-repo
behavior; all other hosts retain the original 2-segment owner/repo split
with remainder as subdir. This preserves backward compatibility for GHE,
Gitea, Gogs, and other non-GitLab platforms while correctly handling
GitLab nested subgroups.
Replace rigid 2-segment HTTPS regex with flexible path parsing that
treats the entire URL path as a repo when no explicit boundary exists.
Use .git suffix as the canonical repo/subdir delimiter, preserving
/-/ (GitLab) and /src/ (Bitbucket) web URL markers.
This fixes install failures for URLs like
onprem.gitlab.internal/org-group/sub1/sub2/project which previously
misinterpreted extra path segments as monorepo subdirectories.
When git operations fail due to SSL certificate issues (self-signed certs,
internal CAs), the error message now suggests GIT_SSL_CAINFO, GIT_SSL_NO_VERIFY,
and SSH as alternatives instead of showing raw git stderr.
Env vars like GIT_SSL_CAINFO are already inherited via os.Environ() in
gitCommand() — the actual gap was error detection and user guidance.
filepath.Walk/WalkDir silently skip symlinked root directories because
os.Lstat sees the root as a non-directory symlink and never enters it.
This affected update, uninstall, reconcile, and server handlers when
the source directory was a symlink (common with dotfiles managers).
Extract ResolveSymlink() from sync.go into internal/utils/path.go and
apply it at every filepath.Walk/WalkDir callsite on source/target dirs.
Also add containment guards to resolveGroupSkills (uninstall) and
resolveGroupUpdatable (update) to reject group directories that are
symlinks pointing outside the source tree.
- Remove duplicated RateLimitError, newGitHubRequest, newGitHubClient,
checkRateLimit from internal/version — now uses internal/github client,
which also fixes missing `gh auth token` fallback for version checks
- Add GH_TOKEN as fallback in install.resolveToken() for GitHub platform,
matching github.GetToken() resolution order (GITHUB_TOKEN → GH_TOKEN)
- Add GH_TOKEN to sanitizeTokens() redaction list
- Extract shared IsAuthError() in install/auth.go; push.go and
wrapGitError() both delegate to it (adds "Access denied" coverage)
- Consolidate shouldFallback*TrackedClone auth checks via isAuthOrAccessError()
- Thread single *http.Client through search.Search() and all callees,
reducing ~60 redundant TLS handshakes to connection-pooled reuse
- Pass *http.Client to downloadFile() instead of creating per-file clients
- Replace sourceHost() with extractHost() (same package, identical purpose)
- Remove unused isGitRepo() alias; callers use IsGitRepo() directly
- Extract shared applyDetailScroll into tui_helpers.go (was duplicated
across diff_tui.go, log_tui.go, audit_tui.go)
- Create utils.FileHash and utils.FileHashFormatted, replacing private
copies in install/meta.go and audit/audit.go
- Export install.MetaFileName constant, use from diff_file.go
- Replace double SHA-256 in fileContentsEqual with streaming byte compare
for early-exit efficiency
- Add 1MB size cap to generateUnifiedDiff to prevent performance issues
- Move noTUI into diffRenderOpts struct, simplifying function signatures
- Fix non-TTY progress bar not incrementing for grouped skills in Phase 2
- Fix Unicode truncation in UpdateTitle using runewidth instead of len()
- Fix IP address regex missing end-of-line bare IPs (add $ anchor)
- Fix gofmt alignment in updateTarget struct comments
- Extract sliceHasName[T] generic helper to deduplicate universal detection
- Add nil meta defensive warning with skipped count in grouped updates
- Replace hand-rolled contains/containsStr with strings.Contains in tests
- Add red team TC-20b for inline code fetch-with-pipe detection
Optimize grouped repo updates by skipping reinstall when installed metadata already matches remote state (commit match, or subdir tree-hash match when HEAD moved).
Emit per-skill grouped progress events so the batch progress bar advances continuously instead of appearing stuck at 0%.
When skipping by tree-hash match, best-effort refresh meta.Version to latest commit to keep status/version output current.
Add unit tests covering commit/tree-hash skip behavior and metadata version refresh.
When update --all processes many skills from the same repo, avoid reinstall/copy when installed metadata already matches the fetched repo state.
For root skills, skip when meta.Version matches repo commit. For subdir skills, require both meta.Version and meta.TreeHash to match. Added unit tests for match/mismatch cases.
UpdateGitIgnore/RemoveFromGitIgnore were called per-skill inside loops,
re-reading the entire .gitignore each time. With large files (100K+ lines
in project mode), this caused O(N×M) complexity and process hangs.
- Add UpdateGitIgnoreBatch and RemoveFromGitIgnoreBatch (single read/write)
- ReconcileProjectSkills: collect entries during walk, batch-update once
- uninstall (global + project): batch-remove after loop, not per-iteration
- Fix spinner WithRemoveWhenDone in StartSpinnerWithSteps
- Add E2E runbook for gitignore batch performance validation
Prevent accidental duplicate installs when the same repo is already
installed at a different path (e.g., --into vs root). Same-repo
reinstalls now show a friendly skip indicator instead of an error.
- Add CheckCrossPathDuplicate() with URL normalization (HTTPS/SSH)
- Add ErrSkipSameRepo sentinel for graceful skip flow control
- Add checkExistingConflict() with repo URL comparison
- Fix --force hint to use repo URL (not per-skill subpath) and
include --into when applicable
- Add dedup check to Web UI install endpoints (HTTP 409)
- Group skipped skills by directory with repo label in summary
Root-level skills (SKILL.md at repo root) have meta.Subdir="" but
discoverSkills() returns Path=".". The mismatch caused lookup failure
in batch update, falsely reporting these skills as stale/deleted.
Switch config install and search install UI from plain spinner to
TreeSpinner with StepResult summary showing installed/skipped/failed
counts and elapsed time. Add git clone progress callback to spinners
so users see real-time cloning status.
- Add Quiet field to InstallOptions for suppressing per-skill output
- Add Skipped field to ConfigInstallResult for accurate summary
- Update integration test assertion to match new summary format
When a registry has multiple skills from the same git repo,
InstallFromConfig now clones the repo once and installs each skill
from the local clone via InstallFromDiscovery. This mirrors the
optimization already present in search_batch.go for hub installs.
The install flow is restructured into 4 phases:
1. Tracked repos (unchanged, each does its own clone+discover)
2. Grouped plain skills (clone once per repo, install from discovery)
3. Singles (unchanged per-skill Install flow)
4. Reconcile config
Skills that fail to match in grouped discovery fall back to Phase 3.
Use a repo-root source in grouped search installs by normalizing from CloneURL instead of carrying subdir Raw values. This prevents malformed metadata sources (for example duplicated subdir paths) and keeps provider-agnostic behavior for GitHub/GitLab/Bitbucket/Azure and SSH URLs.
Also centralize discovery source formatting for metadata so whole-repo SSH installs use //subdir syntax consistently.
Add regression coverage:
- cmd unit tests for grouped clone source normalization across providers
- install unit tests for discovery source formatting (including SSH)
- integration test in tests/integration that verifies grouped install metadata source round-trips through ParseSource.