Agents on a disposable VM, in a hosted assistant or behind an MCP gateway can
reach an MCP endpoint but not the folder skills sync to, so they got tools and
none of the user's skills (#428).
`skillshare mcp serve` is a read-only server for the skills skillshare already
manages. It declares io.modelcontextprotocol/skills and implements skills/list,
skills/get and resources/read: one entry per skill, its URI from the source
path, the full frontmatter, and every file with a sha256 digest and size.
--target applies that target's selection, and the source is reread at most
every 5 seconds so install, update, enable and disable show up without a
restart; unchanged files keep their digest. stdio is the default; --http
serves Streamable HTTP and needs SKILLSHARE_MCP_TOKEN off loopback.
Skills the extension cannot carry are skipped with a warning instead of being
published incomplete: a name that breaks the Agent Skills rules or differs from
its directory, more than 512 files or 16 MiB, or a nested skill that is not
served. --check lists them without serving. The naming rules moved to
internal/skillpkg so sync and the server share them.
Few clients support the extension yet, so the server also offers list_skills
and read_skill tools. A client that declares the extension does not see them,
so it never gets each skill twice.
sync.SelectTargetSkills is the one place that turns discovered skills into a
target's selection (enabled, include/exclude, frontmatter targets).
Refs: #428
The audit names a result relative to the source it was given. With the
runner's 8.3 TEMP the source was the short form and the scan target the
long one, so the name became a chain of ../ segments.
Refs #344
The windows-latest runner's TEMP is an 8.3 short path
(C:\Users\RUNNER~1\...), while the product reports the resolved long
path. Four tests compared the short form and failed only on the runner;
the local guest has no short names. Resolve the expected path first.
Refs #344
These packages were left out of the windows-test job because of failures
that came from test assumptions (POSIX paths, permission bits, HOME,
bash shims posing as git), not product behavior. Fix them and add the
packages to the list so they stay green.
Two failures were real product bugs:
- swapStagedIntoSource named its backup like the temp that
sourcefs.MoveIn creates beside it. On a coarse clock (Windows) both
names matched, the failed move removed the temp and with it the
backup, and the old skill was lost.
- MetadataStore.GetByPath looked up OS-native paths in slash-form keys,
so `update --group` missed nested skills on Windows.
The test sandbox now also sets USERPROFILE, which os.UserHomeDir reads
on Windows; tests were reading the real profile.
Refs #344
`update --all` only consulted the follow policy for `_`-prefixed
directories, so a followed checkout linked under another name (for
example via `link --name dev-skills`) whose root holds a SKILL.md with
recorded metadata was scheduled as a regular skill and `git pull`ed, and
reset under `--force`, against the user's own checkout. Followed links
carry no naming requirement, so run the guard for every directory before
the tracked-repository naming rules apply, in both the global and project
update loops.
Follow-up to the Codex review on #417.
link -h loaded the config first, so it failed with "config not found"
without one and, with --project, could create .skillshare files just to
print help. Detect help among the option arguments before loading the
scope. Pass --name as a value flag to parseModeArgs so "--name -g" no
longer selects global mode (#422).
A followed checkout whose root is a skill is classified as a tracked
repository, so a dirty checkout hit the uncommitted-changes preflight and
was told to retry with --force before the linked-root guard ran. Run the
guard right after target validation for every mode so the answer is always
"use unlink".
POST /api/source-links with enable: true left the new link in the source
when the config could not be saved, so the dialog reported a failure the
filesystem did not match and a retry failed with "already exists". Share
one sourcelink.Discard rollback between the CLI and the handler.
Pass value-taking options to the shared mode parser so command values cannot select a scope or create false mode conflicts. Preserve the end-of-options delimiter and reuse it for MCP command arguments.
Cover existing CLI entry points, including audit rules, enable/disable, and backup deletion, with regression tests and document the parser contract.
The write ratchet flags raw os writes below the skills source; the
rollback added for a failed --enable removed the fresh link with
os.Remove. Go through sourcefs.Unlink so only the link entry goes and
the boundary stays enforced.
link --enable created the link, logged success, and only then wrote the
config; a failed config write left a link discovery ignores behind a
logged success, and a retry failed with "already exists". Enable before
logging and remove the fresh link when enabling fails so the command,
the filesystem, and the operation log agree.
Hub index, built-in search, and draft candidates discovered only the source path, dropping the follow policy used by the inventory. Add optional walk options to the shared Hub functions and pass the operation policy from dashboard handlers and CLI indexing.
Explicit CLI sources use the active config policy while retaining configuration-free indexing. Cover global/project modes, enabled/disabled following, all three handlers, and CLI source overrides; document the behavior in all website locales and built-in guidance.
Standard target naming uses frontmatter names, so matching target entry names misclassified unavailable source skills as broken and suggested pruning. Read the stored link destination, resolve relative paths, and compare it with unavailable source roots; retain name matching only when the link target cannot be read.
Cover absolute and relative destinations and an unrelated broken link whose name resembles a followed skill. Document that standard target names receive the same waiting classification.
Followed worktrees and submodules were discovered as tracked repos but bypassed guards that required a .git directory. Widen IsGitRepo to accept a directory or regular gitdir file and reuse it in discovery and reconciliation. All existing callers classify checkouts or run Git commands; none requires the directory layout, so a separate predicate would retain inconsistent decisions.
Exercise actual followed worktrees in CLI and dashboard update fixtures, including project mode and force retries, and document that they receive the same checkout guards.
Filtered checks dropped the follow policy when resolving a basename, so a linked checkout visible in an unfiltered check was reported as not found by its name. Forward the operation policy to the resolver and cover both global and project checks against a local Git remote.
Document checking followed checkouts by link name in the command reference and built-in guidance.
Missing-repo discovery ignored the operation follow policy, so tracked metadata added a skipped checkout back to update --all and allowed --force to reset it. Thread the policy through missing-repo detection and its CLI/server callers, including rehydration, so existing followed checkouts remain excluded.
Cover global and project updates with dirty tracked checkouts and force, as well as shared and dashboard rehydration. Document that --force does not override the batch skip.
* feat(omp): add native Oh My Pi support
Support OMP skills, instructions, MCP, code hooks and guarded extension selection using its verified native contracts rather than Pi aliases.
Add reviewed marketplace plugin installation, cache-preserving scoped removal and fresh-cache reinstall. Align Pi/OMP extension cards and localize inventory guidance while keeping native ownership and read-only constraints.
Include regression coverage, isolated E2E runbooks, documentation and verification policies. Native plugin updates and unverified runtime platforms remain guarded.
Refs #409
* fix(omp): guard runtime ownership and localize inventory guidance
Native install recursively replaces its runtime destination, and a missing cache manifest could leave the real runtime link installed after removal. Bind and recheck an absent reviewed destination, and recover removal names only from verified scope-local links.
Localize fixed backend reasons and warnings across dashboard locales while preserving paths and external diagnostics. Restore the reviewed source in the existing reinstall fixture and record real native verification.
* fix(omp): allow forgetting natively removed plugins
A different plugin can retain node_modules after native uninstall deletes this installation. When its cache, registration and known runtime state are absent, forget only the stale Skillshare binding instead of demanding runtime-name proof for unrelated files.
Extend the existing removal recovery test with a failing-before-fix case that verifies the binding is dropped and unrelated native lock bytes and plugin files are preserved. make check passes.
Dry runs returned before the linked-root guard, so text preview claimed a trash move and JSON listed a root-skill link as removed even though real execution refused it.
Run the existing non-mutating move check before either preview and extend the root-link fixture to verify text and JSON refusals without claiming removal.
A root skill discovered through a followed source link occupies the link entry itself. Moving that entry to trash silently unlinks every nested skill instead of removing one skill.
Share the root-skill guard across single and batch API handlers and CLI uninstall, report the unlink action clearly, and preserve explicit unlink and nested skill removal.
Stop parsing mode and command options after -- so valid link names and relative paths beginning with a hyphen can be addressed. Document the unlink form in every command locale and verify that -g and -p after the terminator remain names.
follow_source_links only follows a link the user made by hand, so the
guards (source or parent, sync target overlap, link chain) surfaced in
doctor only after the link existed, and Windows users had to pick
between a junction and mklink /D themselves.
link checks the prospective target with Follow.Allow, which shares the
guard rules and reason strings with the walk, before creating a junction
on Windows or an absolute symlink elsewhere. It warns when the target is
not a git checkout and leaves follow_source_links alone unless --enable
is given. unlink moves only the link entry to trash, the same way
uninstall removes a first-level link, so the target is never touched and
trash restore undoes it. internal/sourcelink holds both operations so the
dashboard can call them too.
Refs #419
Keep one follow policy per list and status operation and report its skipped links without hiding healthy inventory. Send JSON diagnostics to stderr and expose sourceLinkWarnings in the dashboard resources response, reusing the sync warning formatter.
Pass the operation follow policy into basename and glob resolution so linked Git checkouts and full skill paths can be updated explicitly. Keep update --all skipping user-owned followed checkouts with its existing warning.
uninstall --group on a followed link is now refused, and restore no
longer creates parents with a raw MkdirAll, so the ratchet allowance
for it is stale.
sync --json showed pruned: 0 and success while a link was unavailable,
so automation could not tell a held-back prune from nothing to prune.
The run's source link warnings now ride along as warnings.
uninstall --group <followed link> moved every skill out of the user's
checkout, while uninstall <link> removes only the link entry. The group
form is refused with a pointer to the by-name forms.
Restore skills through sourcefs with the active global or project follow policy so nested links and disabled following cannot redirect writes or consume the trash copy. Preserve relative link traversal when relocating links, and keep link rows in the trash TUI on their detail view.
update --all treated a followed source link as a managed tracked repo:
with --force it ran git reset --hard inside the user's own checkout and
silently discarded their local edits, and an audit-gate rollback would
do the same. --all now skips any tracked repo behind a followed link
with a warning that names the by-name form, which keeps its documented
behavior.
* fix(diff): compare extension agent outputs and apply target filters
#392 taught status and doctor to recognise converted extension outputs,
but diff (CLI) and the Sync tab (/api/diff) still compared the target
directory's *.md files against the source name. A target whose extension
renames the output (codex -> .toml) therefore listed every agent as
pending after a sync.
Both diff paths now use the same manifest-based check as status, and
report tracked outputs whose source is gone as prune items. They also
apply the target's include/exclude and the agents' frontmatter targets,
so an agent that sync would skip is no longer shown as pending.
Refs #391
* chore(release): request 0.24.6
The pending feat commits would make Release Please propose 0.25.0. This
batch is meant to ship as the 0.24.6 patch, so ask for that version.
Release-As: 0.24.6
An audit run with follow_source_links whose linked checkout was missing
finished as a clean success, so a partial scan looked complete. Audit now
warns once per skipped link in human output and adds warnings plus
incomplete: true to JSON output, including when no skills were found.
Also cover project mode: reconcile keeps config and metadata entries of an
unavailable link, and sync -p warns once and keeps target entries.
A link into a sync target that is not created yet was reported as a
missing target, which marked the inventory incomplete and held back
pruning for a link that would never be followed. The where-it-points
guards now run on the link text before the does-it-exist check, so such
a link is an overlap and the run prunes as usual.
doctor reported the dangling target links behind an unmounted source
link as broken symlinks and suggested `skillshare sync` to prune them,
while sync deliberately keeps them. Those links are now a warning that
names the waiting source link and carries no prune suggestion.
Keep the analyzer metadata context logical while walking the resolved root, so integrity and publisher checks still find the source metadata store. Add global/project command regressions for linked repo roots and single skills, plus linked install-audit and trash sibling-list checks.
Pass the existing optional Follow policy through install audit gates and final metadata hashing, preserving default-off behavior. Absolute link text remains unchanged on trash rename fallback; relative links continue to store their absolute target.
update audited a skill below a followed source link with a scanner that
did not follow the link, and uninstall --group resolved a followed group
without the operation's policy. Both now use the same policy as the rest
of the operation: the update context's follow for the audit scan, and the
uninstall mode's walk for group resolution.
Refs #274
Keep the canonical skills source root and logical relative tail when creating or reformatting project links, including Windows relative links and junction fallback. Shared agent and extra links retain full canonicalization, and global Unix link text stays unchanged.
Allow group walks outside the source only for an enabled first-level link validated by Follow.Resolve, remapping children to logical paths and metadata keys while retaining nested-link guards and root-skipping semantics. Resolve copy and checksum operation roots before walking to avoid false cycles, and preserve followed source links during forced collect with per-item warnings.
Regression coverage verifies second-sync ownership, status, prune, unlink, format changes, linked-root copying, group guards, and forced collect preservation. Linux build, vet, sync and command tests, sourcewalk ratchet, server collect tests, context validation, and Windows cross-build passed; Windows runtime was not verified. No new raw-walk allowance was needed.
Trash moves and restores the link entry without copying or removing its target, including when rename fails. Relative link text is stored as an absolute target so relocation cannot change its meaning; list and TUI expose the link target without inventorying its contents.
Pass the operation-scoped Follow policy into audit and hash operations, resolving their physical root while keeping logical result and metadata keys. Calls without a policy keep their existing behavior and ScanSkill retains its original function signature.
Update six stale trash raw-write allowances for the injected rename test seam and classify moveLink creation/removal: it moves a checked source entry to trash or restores an entry into source, never writes through its target. Raw sourcewalk calls and allowances are unchanged.
With follow_source_links on, discovery reports <source>/_dev-skills/foo,
but every write to it was refused: sourcefs rejects a link in the path, and
its source-root os.Root refuses descendants outside the source. Dashboard
edits, nested uninstall, regular skill replacement, and tracked installs
under such a link all failed, although writing through a link the user
placed is the agreed behavior (#410).
sourcefs.Open, Create, MkdirAllIn, and CheckMoveOut take the operation's
optional sourcewalk.Follow. A clean name <link>/<rest> whose first
component the policy follows is handled by a second os.Root opened at the
link's canonical target, with the same per-component link check on <rest>,
so a nested link or ".." still cannot escape the target. Names with ".."
keep the source root, which refuses them through the link. Without a
policy nothing changes.
The link entry itself stays a link: RemoveAll, Rename, and writes on it are
refused, install --track --force still refuses to replace it, and
CheckMoveOut still lets trash rename the link without following it.
MkdirAll on a followed link is a no-op, since it already is that directory;
install --into needs that. A rename across the link boundary is refused.
The install, update, uninstall, and server callers pass the same policy
their discovery uses, through InstallOptions.SourceFollow and the update
context. No raw-write allowance was needed.
Refs #274
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
Source links can record machine-local paths in Git. Share advisory indexed and ignored checks across CLI, init, and dashboard staging and previews, with scope-correct ignore patterns and manual untracking instructions, without changing staging behavior.
* fix(audit): keep the skills source when auditing agents only
`skillshare audit --agents` replaced sourcePath with the agents source
before building auditTUIContext, so the TUI's Skills tab discovered
skills inside the agents directory. The agents-only branches now read
agentsSourcePath themselves and the skills source is left alone.
Also from the #390 review, independent of .skillfollow:
- /api/skillignore reads the source through skillsSource() like the
other routes (equal in project mode today; consistency only).
- scripts/windows/e2e-file-links.ps1 checked doctor output for a
"Config:" line that no longer exists and could not match the ~-shortened
path, so the run always aborted. It now checks status --json
source.path, as e2e-skillfollow.ps1 does.
- windows-test runs internal/sourcewalk; GOOS=windows vet passes and the
job on this PR is the first real run.
Refs #404, #407.
* fix(audit): relativize agent names against the agents source
With discovery moved to the agents source, auditInstalled and
auditFiltered still printed the header and computed SkillName against
the skills source, so `audit agents --json` reported names such as
`../agents/helper.md` when the two sources are siblings (Codex review on
#412). auditScanRoot now picks the root per kind for discovery, header,
and relativization; the skills source is kept only for TUI tab switching.
Also drops internal/sourcewalk from windows-test again: the first run on
windows-latest fails TestWalkParity, TestWalkDirParity and
TestSymlinkRoot (callback paths come from the canonical root, so
filepath.Rel against the test's root yields `..\..` chains). Logged on
#344.
countAgentLinksAndBroken counted every healthy .md link in the target,
so a leftover link for an agent the target no longer syncs could stand
in for a missing expected one and hide drift. status and doctor now
count a link as linked only when its name is an expected agent; broken
links are still reported regardless of name.
Refs #395, Codex review on #397.
status counted every discovered agent against each target, while sync
agents writes only the agents left after .agentignore, the target's
include/exclude and the frontmatter targets field. A target that leaves
some agents out was reported as N/M with drift after a complete sync,
in the table and in --json, in global and project mode. doctor and the
dashboard already filtered per target.
status now applies the same three filters before counting and expecting.
The table reports an invalid include/exclude as an error, like it does
for skills; --json falls back to the active agents.
Refs #395
An agent target with an extension writes each agent under output_ext
(reviewer.toml) with transformed content, and the manifest tracks that
output name. status, doctor and the dashboard's Sync tab counted synced
agents by looking for the source name and comparing with the source
content, so every converted agent was missing: status showed 0/N with
drift, doctor said the same, and the Sync tab kept them pending after a
successful sync.
For an extension target, count an agent as synced when the manifest
tracks an output with the same stem and that file still matches its
recorded hash. Matching by stem avoids loading the extension spec in
status paths. doctor keeps reporting leftover broken links for these
targets.
Refs #391
Installing a repo whose root is a skill copies the whole repo into the
source, including any .claude/skills/ the author keeps for their own
development. Install discovery already skips target dotdirs, but the
source walk used by sync, check, and the dashboard only skipped .git,
so those dev-only skills surfaced as nested skills and were synced to
every target.
Share the skip set through sourcewalk and apply it in both source
walks. Only dotdirs nested below the source root are skipped: host-style
paths directly under the root (.cursor/skills/*) stay visible because
they drive target inference.
Implements rollout steps 1 and 2 of proposals/274-skillfollow.md. Nothing follows a link or parses .skillfollow yet.
Step 1: doctor reports first-level symlinks and junctions in the skills source as info, in text and JSON, without following them.
Step 2a: internal/sourcewalk owns traversal of the skills source. Thirteen raw walks move onto it with identical behavior, and a go/parser ratchet fails the tests on any new unclassified walk.
Step 2b: internal/sourcefs wraps an os.Root opened at the skills source and refuses any write whose path has a link component. The six write sites the proposal found in review move onto it, including the cross-device copy fallback, and a second ratchet covers raw os write calls. The only behavior change is that an explicit path crossing a link inside the source is refused instead of written through.
Refs #274
Release-As: 0.24.3
The --update-mode help described passive as only pointing facts out,
which dropped the on-request saving that the generated rule, dashboard,
and docs all keep.
Refs Codex review on #384.
The generated guidance block defined "what is worth saving" the same way
agents' built-in memory does, but never said how the two relate, so agents
with their own memory kept saving there and left the shared folder alone.
The active rule now says to save here instead of tool-specific memory,
gives a note format and an INDEX.md line shape, and both modes read
INDEX.md at the start of each task instead of "when relevant". The
passive rule asks agents to point out facts worth keeping so the user
can ask for them to be saved.
Dashboard mode descriptions, the CLI help, and the website pages are
updated to match. Test fixtures that anchored on "Read only" follow the
new wording.