A symlink-mode target folder is the source itself, so it also exposes
skills that .skillignore hides. Doctor, sync, and the dashboard pass
discovery results without those skills, so a symlink-mode claude
target looked harmless to OpenCode even when OpenCode loaded an
ignored skill through it. Treat that pairing as unknown and keep the
previous warning, unless OpenCode links the source as well.
Refs #458
OpenCode loads skill files concurrently, so which copy of a name wins
is not fixed; its own folder does not reliably win. A source skill
synced to both folders is harmless anyway because both copies have
the same content. Count only source skills in OpenCode's folder, so a
local skill there no longer marks a same-named skill from another
folder as already present.
The comparison used each target's filtered selection, which skips the
target_naming: standard validation and collision handling sync applies.
When two skills named dev collide in OpenCode's standard folder, sync
writes neither there, yet a claude target filtered to one of them
still delivered it to OpenCode and no warning appeared. Use the
resolved skills that sync writes instead.
Refs #458
OpenCode reads .claude/skills and .agents/skills first and its own
skills folder last, keeping the last skill of each name. A skill synced
to both opencode and claude therefore loads once, yet doctor and the
sync summary warned on every setup with both targets, with no way to
silence it.
The real effect is different: skills kept out of opencode by targets:
or include/exclude still reach it through another target's folder.
For opencode, count only those skills, name them in doctor, and point
at OpenCode's OPENCODE_DISABLE_CLAUDE_CODE_SKILLS and
OPENCODE_DISABLE_EXTERNAL_SKILLS. Other runtimes keep the old check
until their rule for same-named skills is verified.
Refs #458
FoldHomePathWith writes ~\plug on Windows and a bare ~ for the home
directory itself, but canonicalSource only expanded ~/, so with
preserve_tilde_on_save those sources came back relative on the next
load and update or sync could not resolve them.
When config.yaml is a symlink whose target does not exist yet, as a
dotfiles manager may create before first use, EvalSymlinks fails and
the save renamed over the link again. Follow the link to its intended
target so the first save creates that file, as Config.Save already does.
Refs #460
Plugin saves wrote a temp file next to config.yaml and renamed it over
the path. When config.yaml is a symlink into a dotfiles folder, the
rename replaced the link with a regular file, so the shared config
never received the change and the machine silently forked. The plugin
writer also skipped preserve_tilde_on_save, saving sources as absolute
home paths. The one-time legacy target migrations for global and
project configs used the same temp+rename and replaced links as well.
Plugin sources were compared as raw strings. The snapshot owner marker
records the absolute path from add, so a config that spells the same
folder as ~/... failed every update with "snapshot ownership conflict".
For Agents whose snapshot location hashes the source, the new spelling
silently created a second snapshot instead.
Resolve the symlink before writing, normalize sources once where the
plugin section is decoded so owner checks and snapshot paths agree,
and fold sources back to ~ on save when preserve_tilde_on_save is set.
Refs #460
Refs #461
The CLI and dashboard update a tracked repo through update.TrackedRepo,
which never looked for submodules. A pull that turned a skill directory
into a submodule reported success and the skill vanished on the next
sync. TrackedRepoResult now carries the skipped-submodule warnings and
both callers show them.
Refs #433
CLI search results install a repo-root skill through installFromGit without discovery, so the skipped-submodule warning has to come from that path too.
A failed or timed-out git ls-tree used to look like a repo without submodules, letting an empty submodule path through; the refusal check now reports the failure while warnings stay best effort. Tracked updates also warn about submodules in the pulled tree.
Gitlink paths and .gitmodules URLs come from an untrusted repo, so quote them with %q to keep newlines and terminal escapes out of CLI output. Compare each ancestor of the subdir with the gitlink directory by os.SameFile instead of a byte-length prefix, so case and Unicode-normalization variants that open the same directory are caught.
A --track install with an explicit kind skips discovery, so check the subdir after the tracked clone and remove it on refusal. Match a case variant of a gitlink path only when it opens the same directory, so Linux keeps a real Vendor/Up next to a vendor/up gitlink. Strip everything before the last @ in a submodule URL so malformed and scheme-relative values fail closed.
On Windows and macOS a case variant such as Vendor/Up opens the empty submodule checkout, so compare gitlink paths with EqualFold. Search results that install straight from discovery now show its warnings too.
Redact query, fragment and userinfo from relative, scp-style and malformed .gitmodules URLs too, clean the subdir before matching gitlinks so vendor//up is caught, and return discovery warnings from /api/discover so the install dialog shows skipped submodules.
skillshare never fetches submodules, so a gitlink is an empty directory
after any clone. Installing that path reported success with nothing in
it, a path inside it failed with a misleading "does not exist", and
whole-repo discovery dropped its skills without a word.
Refuse a subdir that is or lies inside a submodule, naming the
submodule, its pinned commit and upstream URL, and warn for each
submodule skipped by whole-repo discovery and --track.
Refs #433
Release-As: 0.25.1
The bar is position: fixed with left: 50%, so its shrink-to-fit width is
capped at the space right of its left edge, half the viewport. Content
wider than that wrapped inside the fixed 46px height, as seen on the Pi
and OMP pending-changes bars. Size it to its content and cap it at the
viewport instead.
The pages added with the block modes covered what they write, not what
the review fixes made them do. The sharing guide still said sync
overwrites a modified target, which is wrong for a block: sync stops
until the edit is collected or the block is rewritten.
The guide, the extras reference and the built-in skill now cover the
block's Collect and Rewrite choices, block statuses, the target file
dialog, what a mode change removes, how restore treats a block, and
the cases a block target is refused. Common errors gains an Extras
section for the four new messages. The --mode help of extras init,
--add-target and the mode command listed only import among the
single-file modes; it now lists prepend and append too.
Refs #436
Two gaps found by the fifth Codex review of #451:
- A source line that reads as a block marker, such as an unfenced
end marker, was copied into the body, so the freshly written target
parsed as damaged and every later sync, collect and detach failed.
The block is now parsed back before writing, and a source whose block
would not read back as itself is refused.
- When a block had created the target and the user then added lines
outside it, switching to symlink or copy kept the "created" record,
so restoring the link later removed the file and the user's lines
with it. Those lines are now recorded as the restore point.
Refs #436
Four gaps found by the fourth Codex review of #451:
- The check that refuses a target linked to another extra's source ran
for import only, so a prepend or append target linked that way had
its block written straight into that source. It now runs for every
mode that edits the target in place.
- Switching a target with a damaged block to symlink or copy recorded
the damaged block as the user's restore point, so detaching later
brought it back. Recording the restore point now fails on a damaged
block.
- Appending to a file that ends inside an open code fence put the begin
marker inside that fence, so the next sync missed the block and
appended another. Such an append is refused. Separately, a block body
with an unclosed fence now ends at the end marker whose body matches
the block's hash, even when the user's own code follows.
- A second prepend block counted its sibling above it as misplacing it,
so each sync moved one ahead of the other. Placement now ignores other
managed blocks on the same side.
Refs #436
DELETE /api/repos/{name} now refuses a repo with uncommitted changes or
unreadable git status, and the Dashboard page never sent force, so those
repos could not be removed from it at all.
The two refusals force overrides now carry their own error code. The
page still sends the first request without force; on either code the
confirmation says why and retries with force only when the user
confirms. Other refusals, such as a linked folder that is itself a
skill, keep the plain error because force does not change them.
Refs #442
On main these two readers scanned SKILL.md line by line and returned at
the line they were looking for, keeping nothing. After the refactor they
first collected the whole frontmatter block, so a large block cost its
full size on every call even when name came first, and an unclosed one
cost the rest of the file. Sync, target naming and listing call them
once per skill or agent. Reported by Codex review on #457.
The per-line rule (which line opens and which closes the block) moves
into blockWalk, which locateFrontmatter and the new scanLenientBlock
both use, so the streaming path and the byte locator cannot drift.
ParseSkillName and ParseFrontmatterField stop at their key again; the
list and fields readers collect the block as they always did. Results
are unchanged for every entry point.
Refs #449
Five gaps found by the third Codex review of #451:
- Switching a target with a damaged block (a deleted end marker) to
import ignored the parse error and added the @ line, so the tool read
the stale block as well. The switch now fails until the block is
repaired.
- Detaching such a target reported success without touching the file,
so Assign dropped the connection and left the block with nothing
configured to remove it. Detach now returns the error.
- PlanAssign decided whether a non-import target could take several
files, and Assign chose each new file's mode, from the extras before
the files being detached were gone. Removing the only block holder
while adding two files therefore passed the plan, then gave both new
files a link, the second replacing the first. The plan now detaches
first and Assign applies the modes the plan validated.
- When the last block left a file it had created but the user had since
added lines to, the created record stayed, so a later link attach and
restore deleted the user's file. The attach record is cleared once no
managed block or import line remains.
- A source with a fence that never closes swallowed the block's own end
marker on the next parse, so a freshly written target read as
damaged. An open block's end marker now ends it when the fence never
closes.
Refs #436
TestGitRemoteHash_DefaultBranchUsesAuth puts an extensionless shell
script named git first on PATH. On Windows exec.LookPath only resolves
names with a PATHEXT extension, so the script is ignored and the real
git.exe probes github.com, which answers "repository not found" for the
made-up repository. The auth env is built without any platform branch,
so the product path is unaffected; only the test double cannot work
there. Skip it for the same reason the fake-git tests in
internal/install/source_fallback_test.go already do.
Refs #444
Rebuilding the path readers on the shared locator made them read the
whole SKILL.md into memory, where they used to scan line by line and
stop at the closing delimiter. They run once per skill or agent during
sync and listing, so a large body, or a large file with no frontmatter
at all, now cost its full size on every call.
readHead goes back to a bufio.Scanner and keeps the lines from the
opening delimiter to the closing one, nothing before and nothing after.
Reading through the real Scanner also replaces the code that imitated
its quirks (64 KiB line cap, one \r dropped per line); ReadSkillBody,
which needs the body, takes them from scanLines. An unclosed block is
still kept to the end of the file, as the list and fields readers always
did.
Review follow-ups in the same files: RewriteFrontmatterList no longer
copies the content to a string and back, ParseFrontmatterField's two
loops no longer share an index, and the tests share one file helper.
Results are unchanged for every entry point.
Refs #449
Review showed three ways a missing key still passed:
- t('sync.changes') passed because the counted base was accepted for
any string; a string passed straight to t() must now be a key itself.
- plural('typo.key', n) was only seen when its first word was a known
namespace; strings passed straight to plural() are now checked for
both forms.
- The Go test listed the values that complete "plugins.problem." by
hand, so a new case in automationProblem passed; the values now come
from that function's switch.
The files of config paths were also skipped by every check; they are
now skipped only where config paths look like keys, so their t() calls
are checked again.
Refs #448
Two places still wrote out 'translate the backend key, else show its
text' by hand next to keyedText, so the rule had three copies. The Pi
package test also rebuilt the set of Pi targets for every package on
every render; it is now built once per inventory, as before the flow
moved out of the page.
Refs #446
The audit query of a skill is keyed by name and kind, and its cache is
cleared by the ['audit', 'skill'] prefix. The name-only key lost its last
caller when the key literals moved into queryKeys.ts, and read as the real
query key while matching none exactly.
Refs #445
The mode picker stores the new mode before syncing, so the block and
import handlers saw the previous mode's output as the user's content:
copy → prepend kept the whole copied source beside the new block,
import → prepend kept the @ line, and prepend → import kept the block.
The tool then read the instructions twice, and detaching removed only
the new representation.
Entering a block mode now rebuilds a leftover copy from the attach-time
base (backing up an edited copy as a mode-switch drift, as import does)
and drops this extra's @ line; entering import mode drops this extra's
block, keeping a hand-edited one as a drift backup.
Refs #436
Review of the shared uninstall operation found that RemoveByNames still
matched a bare key: removing top-level "foo" pruned the legacy entry
"foo" with Group "frontend", which belongs to the installed
frontend/foo. Entries are now matched by their full path only.
Also from review: a dry run no longer runs git status for tracked
repos (it reports no git state), the unused Outcome.Removed is gone,
the two single-item handlers share one status mapping, and the CLI
reports a single target in one place instead of three flag checks.
Refs #442
The dashboard update of N tracked repos loaded the metadata store 2N times
under the server lock, and each load walks the whole source tree. The refresh
now returns before loading when the repo has no root SKILL.md, reports
whether it saved a change, and the server reloads its store only then,
through the existing reloadSkillsStore.
The CLI single-repo path tracked its output with two flags and a OnceFunc.
The caller hook now runs once for every outcome, so the spinner stop, the
discard warning and the pull summary live in one place and print in the same
order as before.
auditGateFailClosed rebuilt three of its messages by hand; they are
AuditGateError's own wording with another subject, so both now come from one
function.
Refs #443
The source view of a target's file showed the block markers as plain
comments, so which lines skillshare manages, and which block had been
edited by hand, took reading the markers. Each managed block is now
tinted, a block the target reports as modified in the warning tone, and
the dialog says what the block is or offers Collect and Reapply for the
edited one, as the row does.
Refs #436