1349 Commits
Author SHA1 Message Date
Willie 52744e2f34 feat(project): register every Git checkout in a workspace with --scan (#958)
Some tools, such as Codex, look for project skills only up to the Git root
of the folder you open. When one workspace folder holds several checkouts
or worktrees, skills synced to the workspace are not seen inside them, and
registering each checkout by hand repeats the same settings many times.

`project add <root> --scan DIR` registers the root and each Git checkout
found below the scan folders as ordinary project entries with the same
settings, in one write. No scan setting is saved; rerun it after a new
clone. It only adds: a checkout already registered with the same settings
is kept, one with different settings or its own .skillshare/ is skipped
with the reason, and a name clash derives the name from the path. A scan
folder that is missing or cannot be read is skipped with a warning so one
unreadable folder does not block the rest. Rows carry a root-relative path
and a skip code so the dashboard can explain them in its own words.

Dashboard:

The dashboard should offer what `project add --scan` does, so a workspace
with many checkouts can be set up without the terminal.

The Add project dialog gains "Also add each checkout inside this folder"
with the folders to look in. Review runs the scan as a dry run through the
new POST /api/projects/scan and lists each folder as Added, Kept or Skipped
with a translated reason and a link to a project whose settings differ;
nothing is written until it is confirmed. The write scans again and counts
what it actually added, so a checkout registered in the meantime is not
reported as new. MCP and Hooks, when picked, are set up for each added
project. Without an explicit storage choice the default follows
projects_storage, as the CLI does without a flag.

Closes #953

Release-As: 0.30.12
2026-11-19 01:35:56 +08:00
Willie f8d768a256 fix(projects): say where a removed project's MCP and hook settings go (#955)
Removing a project keeps its mcp.projects and hooks.projects entries on
purpose: they are separate lists that no project command edits, and the
dashboard lists such a folder on its own. But nothing said so. The CLI
warned "mcp.projects still contains root X" with no next step, and the
dashboard dialog only mentioned it for a machine-local project, so a
removed project's MCP settings looked like a cleanup bug.

The CLI warning now names the settings and where to remove them. The
remove dialog says they stay and that removing the leftover row deletes
them, except when only this machine's copy goes and config.yaml's
project remains. project.md, ui.md and the skill reference describe it.
2026-11-18 22:10:07 +08:00
Rafael Bodill ef072113a9 fix(plugins): make OMP removal previews deterministic (#949)
Apply rejected its own preview with "plugin state changed since preview"
when an OMP removal ran while other OMP plugins were registered: their
package files were fingerprinted in map iteration order.
2026-11-18 10:16:27 +08:00
Willie 6869c39d16 feat(hooks): edit and remove unmanaged Antigravity and Copilot hooks in place (#948)
The "Hooks in targets" dialog could change or remove a hook without
importing it only when the Agent keeps hooks as events of one settings
file. Antigravity blocks and Copilot hook files had to be imported first,
even though both are plain configuration the dialog can address exactly.

Candidates now carry the block or file they were read from (source), and
NativeEdit uses it: an Antigravity edit replaces or removes the whole named
block, keeping enabled: false; a Copilot edit rewrites only the file's
hooks key or deletes the file. Both keep the preview diff, the stale-file
check, the backup, and the hook stays unmanaged. Code hooks (Pi, Oh My Pi,
Amp, OpenCode) still need importing.

Dialog rows are keyed by source, so two Copilot files that read as the same
hook name (lint.json, skillshare-lint.json) can no longer be removed
together. Copilot rows show the file name, the editor names the file, and
removal says the file is deleted.
2026-11-18 02:10:49 +08:00
ed3c825c49 feat(hooks): record Codex hook trust on sync with --trust (#829)
* feat(hooks): record Codex hook trust on sync with --trust

Codex runs a user-level hook only after its trusted_hash is stored in that
home's config.toml, keyed by the hook's position in hooks.json. Only the
TUI's /hooks review writes it, so every hook Skillshare adds or moves starts
untrusted and codex exec skips it without a word.

`hooks sync --trust` (global mode, opt-in) asks each Codex home's
`codex app-server` for the current hash of the hooks Skillshare owns and
writes trusted_hash through config/batchWrite, the same calls /hooks makes,
so the hash is always the one Codex computes. It changes nothing else in
config.toml: disabled hooks stay disabled and are reported. Trust it wrote
is remembered in the hooks state directory, so when a hook moves or is
removed the next --trust sync clears the old position, unless that trust
has changed since. Without codex on PATH the hook files are still written
and the command fails saying trust was not recorded.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(hooks): scope Codex trust to config owner

Scope trust selection and saved hashes to the active config owner so a global config cannot trust or reconcile another config’s hooks.

Recheck hooks.json after hooks/list to avoid trusting a handler from a stale position, and only report trusted or cleared actions after config/batchWrite succeeds. Keep v1 ownerless hashes cleanup-only so upgraded installs clear stale trust after hook removal without trusting foreign hooks.

* fix(hooks): retain parked Codex trust receipts

Locate hooks using all records for their canonical file before selecting the active config owner, so identical hooks remain identifiable across shared and symlinked homes. Keep cleanup receipts for parked registrations whose native group still matches, including when duplicate counts make locate ambiguous, while clearing receipts for removed hooks.

Add regressions for identical hooks across config owners, symlink aliases, parked-home restoration, ambiguous duplicates, and stale receipt cleanup.

* fix(hooks): coalesce symlinked Codex trust homes

Canonicalize current-owner and legacy trust receipts so symlink spellings of one hooks.json are processed once. Remove the redundant parked-position branch because the hash-matching fallback below covers it. Add regression coverage for current and legacy receipts.

* fix(hooks): coalesce removed Codex hook homes

Resolve existing home directories when hook files have been removed so trust receipts for symlink aliases reconcile once. Correct canonical key expectations and exercise coalescing without unsupported ledger migration.

* fix(hooks): read reconciled trust fixture into fresh state

JSON decoding retains existing map entries, so reusing the seeded receipt map makes the alias appear to survive reconciliation. Decode the persisted state into a new value to verify its actual contents.

* feat(hooks): confirm Codex trust before recording it and drop the receipts file

`hooks sync --trust` recorded trust for every owned hook without showing
what it was trusting, and tracked its own writes in codex-trust.json so
it could clear them later. Codex's review exists so that a person sees a
command before it runs, and the receipts file was the source of most of
the edge cases found in review.

Trust is now two steps. A preview asks Codex which owned hooks are new
or changed and returns them with their commands and a revision; the
write takes that revision and is refused when the hooks changed since.
The CLI lists the hooks and asks once, `--yes` skips the question, and
without a terminal nothing is trusted unless `--yes` is given. The
dashboard gets the same two steps as API routes.

Nothing is cleared any more: a leftover entry only matches a hook with
the same event, matcher and handler at that position, which is the hook
the user already approved. That removes codex-trust.json.

The app-server is started without CODEX_HOME for the default home.
Codex resolves the symlinks of CODEX_HOME but not of ~/.codex, so a
symlinked default home got a trust key the user's own Codex never looks
up.

A home whose write fails no longer stops the other homes, and the error
names which were recorded. A preview releases the hooks operation lock
before it waits on Codex, so a slow Codex does not fail other hooks
operations.

Refs #828

* feat(ui): trust Codex hooks from the Hooks page

The dashboard could sync Codex hooks but left them untrusted, so Codex
skipped them until the user ran the CLI or approved each one in Codex.

The Sync box now offers "Trust Codex hooks" in the global scope when a
hook is bound to a Codex target. The dialog shows each hook that needs
trust with its full command and records trust for exactly that list; a
change since the list was shown is refused. The button is hidden while
changes are pending, because trust covers the hooks on disk and the next
sync would invalidate it.

Refs #828

* test(ui): cover the Codex trust dialog's refused, empty and disabled states

Only the happy path was tested, so a regression in the states where the
dialog must not offer a write would have passed. The docs sentence on
several Codex homes now says what is isolated: a failed write, not a
home that cannot be listed.

Refs #828

* refactor(hooks): split the trust prompt and output rendering out of the handlers

hooks_handlers.go had grown past 500 lines and mixed orchestration with
the confirmation prompt and every print function, which the CLI layering
rule asks to keep apart. The code is moved unchanged: the prompt goes to
hooks_prompt.go and the printing to hooks_render.go.

* fix(hooks): leave out a Codex home that cannot be listed instead of stopping

With several Codex homes, one home Codex could not list aborted the whole
trust run, so a broken account home kept every other home untrusted.

A home that fails to list is now reported and left out, and the other
homes are previewed and trusted as usual. The CLI warns before the
question and exits with an error that names the home; the dashboard
dialog shows the same note. When no home can be listed the preview still
fails.

Codex answers a home it cannot load, such as one with an invalid
config.toml, with a successful reply that has no hooks and the reason in
`errors`. That was read as a home without hooks and passed silently. It
is now a failed listing that carries Codex's own message.

Refs #828

* fix(hooks): check the reviewed Codex hooks per home when recording trust

The revision was one digest over every home's hooks. When a home was
listed in the preview but failed to list again at the write, its hooks
dropped out of the digest, so the write was refused as stale: the homes
the user had just reviewed stayed untrusted and the real listing error
was hidden behind "changed since they were listed".

The revision now has one part per home. A home is refused only when its
own hooks changed; a home that cannot be listed any more is reported and
the reviewed homes are still trusted.

Refs #828

* chore: release as 0.30.10

Keep the next release on the patch version the maintainer asked for,
instead of the 0.31.0 that Release Please derives from the feat commit.

Release-As: 0.30.10

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Willie <egg358358@gmail.com>
2026-11-17 15:54:20 +08:00
Willie c626c4bf1f fix(sync): let --force take over an agent file that already matches (#944)
Without file-link privilege (Windows without Developer Mode) merge mode
copies agents and tracks the copies it wrote. A file a tool already had
with the same bytes as the source was not tracked, so it was kept, and
--force kept it too: the identical-content check ran before the force
check. init's "kept local agents: X (sync agents --force replaces them)"
then named a file --force could never take over, on every run.

With --force the file is now recorded as skillshare's, so later syncs
update it. Without --force nothing changes.

Also: the five-name cap written twice is one helper, sync.ListNames,
and doctor.md says when a recorded broken link is reported as a broken
symlink (the behavior from #942).
2026-11-17 14:01:15 +08:00
Willie 7af9224896 fix(diff): list only what sync prunes, and have doctor follow it (#942)
Three previews disagreed with what sync does to a link no skill or
agent maps to.

The agent diff, in the CLI and the dashboard, listed every orphan link
as a removal. Agent sync prunes a link only in merge mode, and only when
it is broken or points into the source, so a live link to another
location was listed forever. Both diffs now ask the function prune uses.

The CLI diff listed no orphan links at all for merge-mode skills, so a
skill removed from the source showed nothing until sync pruned its
link. It now lists the links sync.OrphanLink says prune removes.

doctor called a broken link "did not create, kept" whenever OwnsLink was
false. Prune also removes a broken link under a name the manifest
records in any form once no skill maps to it, so doctor told the user to
delete by hand a link that sync removes. It now reports those as broken
and suggests sync. An empty source counts as "every entry is an orphan";
only a failed discovery turns the check off.

Closes #939
Closes #940
Closes #941
2026-11-17 13:38:31 +08:00
Willie 5b249b60fe refactor(sync): share one orphan-link rule between prune and the dashboard (#937)
The Sync page predicted which orphan links sync would remove with its
own copy of prune's rule. The copy had already drifted once (#914), and
it still disagreed in two cases: prune treats a link it cannot follow,
such as a loop, as broken, while the page only counted a missing
destination; and prune keeps a link whose target it cannot read.

The decision now lives in sync.OrphanLink. Prune and the dashboard diff
both call it, so a later change to the rule reaches both. Prune's
results, warnings and dry-run text are unchanged. LinksIntoSource, added
for the dashboard's copy, is no longer needed and is removed.
2026-11-17 13:03:42 +08:00
1419b62499 fix(sync): never delete or re-point entries skillshare did not create (#914)
* fix(sync): never delete or re-point entries skillshare did not create

On Windows, a plain `init` followed by `sync` deleted things in
~/.claude/skills that skillshare never created, with no backup and no
prompt. It also re-pointed another tool's junction. Seen in a sandbox
run of v0.30.7:

- prune removed real folders whose names only look like skillshare's
  flat naming (team__shared, _drafts), via the name heuristic
- prune removed a broken link that another tool or the user made
- merge sync deleted and re-created a same-name link that pointed into
  another manager's store
- init's import deleted original folders that matched the source
  byte for byte

The manifest already records what skillshare made, so use it:

- prune keeps a look-alike folder that is not in the manifest and warns
  once per sync; --force still removes it
- prune removes a broken external link only if the manifest records it
  (e.g. after the source moved); otherwise it keeps the link and warns
- merge sync re-points a link only if the manifest records it or it
  already points into the source; otherwise it reports the entry as
  kept local, as it already does for a real folder; --force replaces it
- init's import moves identical copies to the trash under
  replaced-by-init/<target>/ instead of deleting them

Tests: ownership_test.go adds regression tests that use createLink, so
they run on Windows without Developer Mode (junctions). The 4 keep
tests fail without this change. TestSyncTargetMerge_FixesBrokenLink now
records the link in the manifest first, since a same-name link that
skillshare did not make is no longer re-pointed. Existing tests in
internal/sync, internal/trash and cmd/skillshare show the same failures
on a Windows host with and without the change (symlink privilege). The
devcontainer suite has not been run.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* docs: say what sync and init keep when skillshare did not create it

The previous commit makes sync and init leave alone entries skillshare
did not create. Update the pages that describe that behaviour:

- sync-modes: merge mode keeps a same-name link from another tool and
  reports it as local; orphan cleanup keeps folders that only look like
  skillshare's flat names, and broken links outside the source that the
  manifest does not record. sync --force still removes them.
- init: identical folders replaced with links during the first sync go
  to the trash first (replaced-by-init/<tool>/, cleaned after 7 days).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test(sourcefs): drop the raw-write allowance replaceIdenticalCopies no longer needs

replaceIdenticalCopies now moves an identical copy to the trash instead
of calling os.RemoveAll, so TestRawWriteRatchet reports its allowance as
stale.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(sync): apply the ownership rule at every consumer

Review of the first commit found places that still did, or predicted,
the old behaviour. One rule now decides whether a link is skillshare's:
Manifest.OwnsLink (the manifest records a link there, or it points into
the source). It compares canonical paths, so a source reached through an
alias and a relative project link still count as skillshare's.

- copy mode kept re-pointing: it removed any same-name link and wrote a
  copy. It now keeps a link it does not own, like merge mode.
- the dashboard diff predicted a prune for a folder that is only named
  like skillshare's, and an update for another tool's link. It now
  reports them as local and kept. The orphan-link prediction uses the
  same rule as prune.
- `diff` reported a target as in sync while sync printed "kept local"
  for another tool's link. It lists the link as a local override.
- doctor reported a broken link skillshare did not create as an error
  and suggested `skillshare sync`, which now keeps it. It is a warning
  with its own row, and `broken_symlinks` is `warning` in --json.
- `trash restore` put a folder init had replaced into the source as a
  new nested skill. Restore refuses entries under replaced-by-init/ and
  names the trash folder to copy from.
- init reported a folder it could not move to the trash as one that
  differs from the source, and dropped the error. It is reported
  separately, with the error.
- the PruneOptions.ManagedOnly comment still said broken external links
  are skipped; manifest-recorded ones are removed.

Tests cover the states the review listed: a source behind an alias, a
relative project link, ManagedOnly, standard naming, dry run with
--force, and a recorded copy the user replaced with a link.
TestSync_Pruning_RemovesBrokenExternalSymlinks passed without testing
anything, because FileExists follows links; it is now two tests that
use Lstat, one per owner.

Agent merge is unchanged: agents have no link manifest.

* fix(diff): list another tool's link as kept in copy mode too

Copy mode now keeps a same-name link skillshare did not create, but both
diffs read the entry with os.Stat. A broken link showed as a skill to
add on every run, and a live one as a plain local copy, while sync kept
both and said "kept local".

`diff` and the dashboard diff check the link first and report it the way
merge mode does: a local override, kept until --force.

* test(sync): make the aliased-source link while its folder exists

TestSyncMergeRepointsUnrecordedLinkIntoAliasedSource failed on Windows:
it linked straight to a missing folder, and a junction cannot be made
to one. It now links first and removes the folder after, like the other
broken-link tests in this file.

* fix(sync): compare a broken link by where it would point

linksIntoSource resolved both paths with EvalSymlinks, which gives up on
a path that does not exist. A broken link is exactly that, so a link
into a source reached through an alias (a symlinked config folder, or a
short 8.3 path on Windows) compared an unresolved path with a resolved
one and was taken for another tool's link. Sync then kept it instead of
repairing or pruning it.

It now uses sourcewalk.Canonical, which resolves the nearest folder that
exists and keeps the missing tail. The test covers both spellings: the
alias in the link, and the alias in the configured source.

* fix(init): report what the first sync kept, and match prune in the diff

Two previews said less, or more, than sync does.

init ran its first sync quietly and printed only "Synced". Sync now keeps
a link another tool put on a skill's name, a folder that is only named
like skillshare's, and a broken link it did not create, so the skill was
reported as synced while the tool still loaded the other copy. firstSync
returns the warnings a plain sync prints, and init shows them. They
already name the folders that differ from the source, so the separate
list of those is gone.

The dashboard diff counted an orphan link as a removal whenever the
manifest recorded it. Prune keeps a recorded link that still resolves
outside the source (the source moved and the old checkout stayed), so
the Sync page showed a removal that never happened. The diff now uses
prune's own rule: a link into the source goes when broken or recorded;
any other link only when broken and recorded.

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Willie <egg358358@gmail.com>
2026-11-17 04:15:04 +08:00
Willie 310478c28d fix(ui): fetch each repository once when the dashboard updates skills (#931)
* fix(ui): fetch each repository once when the dashboard updates skills

Updating many skills from the Updates tab cloned the repository once per
skill and reinstalled every one, so 31 skills from 8 repositories took over
a minute where `skillshare update --all` takes 16 seconds, and unchanged
skills were reported as updated.

The update stream and the update-all endpoint now group the request's
skills by repository and branch and go through install.UpdateSkillsFromRepo,
as the CLI does. A skill that already matches upstream is left alone and
reported as up to date; the Updates tab shows and counts it that way.
Results still arrive in request order, which the dashboard relies on.

Updating a single skill and switching its branch still reinstall. Those
paths now also pass the configured parse options to the installer.

* fix(ui): keep checkouts and recorded sources when updates share a fetch

Grouping a dashboard update by repository sent every skill through the
discovery installer. A root skill kept as a Git checkout was replaced by
a plain copy, losing its repository, and an updated skill had its source
recorded in the repository's canonical spelling, which a project
declaration written with another spelling no longer matches.

A checkout now updates on its own by pulling, as before, and an updated
skill keeps the source its record had, as updating one skill does.
2026-11-17 02:43:57 +08:00
Willie 0dc2018747 feat(hooks): browse, edit and remove unmanaged hooks in the import dialog (#920)
* feat(hooks): browse, edit and remove unmanaged hooks in the import dialog

The import dialog listed every target's hooks in one long scroll of
full commands, so it was hard to read and hard to tell which target a
row belonged to. It also offered no way to change a hook without first
handing it to skillshare.

The dialog now shows one target at a time from a rail, with two-line
rows that open to the full commands. A hook that lives as an event in a
target's settings file can be edited or removed in place: the change
stays pending, a review step shows the file diff, and only then is the
file written, with a backup that restores it. The hook stays unmanaged.

The write is a new operation in internal/hooks (PreviewNativeEdits and
ApplyNativeEdits) that addresses elements by content, refuses anything
a ledger record holds, and reuses the existing backup and journal path.
Whole-file hooks, Copilot files and Antigravity blocks still have to be
imported first.

* fix(hooks): keep repeated restores and retries from writing a hook twice

Two ways an unmanaged hook could end up in a target file twice:

Restoring the backup of a removal a second time appended the hook
again. An unmanaged hook has no ledger record, so nothing marked the
backup as restored. PreviewRestore now reports a conflict when the
removed hook is already in the file.

When the dialog's native write succeeded and a later import failed, the
retry compared each pending change with the row it was made in. An edit
that moves or adds a command under another event comes back as other
rows, so the change stayed pending and was written again, and the new
row was picked for import. The dialog now drops every change once the
write went through (or reports files it already applied), and keeps
only the rows picked before as imports for the targets it wrote.

* chore: release as 0.30.9

Keep the next release on the patch version the maintainer asked for,
instead of the 0.31.0 that Release Please derives from the feat commit.

Release-As: 0.30.9

* fix(hooks): tell a restored hook from a surviving twin, keep unwritten edits

The check that refuses a second restore of a removed unmanaged hook
looked for any unmanaged hook with the same content. When one of two
identical hooks was removed, the survivor made the first restore look
done. A removal that shifts later hooks is stored as an update plus a
removal, which the check did not cover, and a restored hook that was
imported afterwards no longer counted, so those backups could still
restore twice. Every change to an unmanaged hook now records how many
identical hooks the write left, and a restore is refused only when the
file holds more than that.

When a native write failed after its first file, the dialog dropped
every pending change, including those of files the write never reached.
The failure response now lists the applied files in error_params, and
the dialog drops only the changes of those files. Pasted candidates are
no longer unpicked by a write to the target they were pasted for.

* fix(ui): show a hook row's edit and remove actions once

An open row in the import dialog showed the pencil and trash icons next
to the labelled Edit and Remove buttons below, which do the same thing.
The icons now hide while the row is open; their space is kept so the
chevron stays where it was clicked.

* chore: release as 0.30.9

Keep the next release on the patch version the maintainer asked for,
instead of the 0.31.0 that Release Please derives from the feat commit.

Release-As: 0.30.9

* fix(hooks): count surviving twins for every backup, not only native edits

Only the native edit preview recorded how many identical hooks a write
left, so the backup a restore writes had no count. Undoing a restore
through Backups was then refused whenever a hook with the same content
was still in the file. The count now happens where every backup is
written.

* fix(ui): keep the row icons as the only edit and remove actions

The labelled Edit and Remove buttons under an open row repeated the
icons and stood out too much, so they are gone and the icons stay
visible on open rows again. Each icon now has a tooltip that says what
it does.

* chore: release as 0.30.9

Keep the next release on the patch version the maintainer asked for,
instead of the 0.31.0 that Release Please derives from the feat commit.

Release-As: 0.30.9

* fix(ui): put the import dialog's hooks on one card, with fewer lines and fills

The dialog mixed a grey rail, a grey footer, grey expanded rows and a
divider around every region, and the editor stacked its own Cancel and
Done above the dialog's. It read as cluttered, and a long JSON line
pushed the editor's controls out of the dialog.

The dialog now sits on the page background with the hook list, or the
editor, as one card. Rows are separated only by the list separator, the
row icons are neutral until hovered, and code is the only sunken fill.
One row is open at a time, and a single command no longer repeats its
event as a heading.

While a hook is edited, the dialog's footer holds the editor's actions,
so there is one row of buttons and Cancel returns to the list instead
of closing the dialog with the draft. Commands are sections of the card
and wrap instead of scrolling sideways. The dialog is larger.

The hint is one line, the edit note moved to the footer, and the name
field is labelled by where the name is used.

* chore: release as 0.30.9

Keep the next release on the patch version the maintainer asked for,
instead of the 0.31.0 that Release Please derives from the feat commit.

Release-As: 0.30.9

* fix(ui): give the import review step the same card-on-ground look

The review step still had the white body, the grey footer and the
bordered summary box, so it looked like a different dialog from the
list it follows. It now sits on the page background with the summary
as one card, and the removal tile is neutral with only its icon in the
danger colour.

* chore: release as 0.30.9

Keep the next release on the patch version the maintainer asked for,
instead of the 0.31.0 that Release Please derives from the feat commit.

Release-As: 0.30.9
2026-11-16 22:55:13 +08:00
Willie 94cf438066 fix(audit): match only uppercase role labels in source files (#923)
prompt-injection-1 kept reporting ordinary source lines as CRITICAL
prompt overrides: `root: Path = Path(".")` (#921), docstring arguments
such as `root: Repository root containing the manifests.`, and typed
parameters with a name default. #376 and #750 each added value shapes
to the exclude. Measured on 5,573 real Python files and 33,006 Markdown
and config files, widening the exclude again cleared no line, and a
docstring entry cannot be told from prose by any value pattern.

In a source file a lowercase label word is a parameter, field or
docstring entry, so the built-in rule now matches only uppercase labels
there. The file extension decides, never a shebang, so a text file
cannot switch the check off. Markdown, text, config and extensionless
files, and Markdown code fences, keep the case-insensitive match. A
custom regex under the same ID is used as written.

This reverses the earlier choice that a lowercase `system: "..."` in a
non-Markdown source file stays CRITICAL.

Closes #921
2026-11-16 22:13:30 +08:00
Willie 5a6873ec96 fix(mcp): stop conflict resolution looping between Agents (#919)
When Agents hold different definitions of one server, importing one
Agent's version was refused with "unresolved conflicts prevented
adoption", because any conflict in the plan blocked it, including
another Agent's. The source was saved anyway, so the Agents that had
matched the old definition now conflicted, and importing from them
flipped the source back. Replace with source sent one resolution at a
time, so with two conflicts the plan stayed blocked and the dialog
could never apply.

Gate adoption only on the resolved entries' own conflicts, in the API
save and in the interactive CLI import. Replace with source now settles
every conflict of that server at once. After a conflict import, a
translated hint names the Agents in the same scope that still need
Replace with source, and the remaining adoption error is translated.
2026-11-16 15:48:20 +08:00
Willie ffbddb608a chore(audit): name the SARIF rule index key correctly
The map has been keyed by ruleID, with Pattern only as a fallback,
since rule IDs were introduced; the comment still said pattern.
2026-11-16 13:54:03 +08:00
Rusty Shackleford 22fdb28de0 fix(audit): report maximum SARIF descriptor severity (#916)
Python credential scans can produce a HIGH comment before a CRITICAL
code finding for the same rule. Keeping the first descriptor score hides
the later severity. Promote the shared descriptor while retaining each
result's level and the first-seen rule index.
2026-11-16 13:53:45 +08:00
Rusty ShacklefordandWillie 43059a1428 fix(opencode): preserve v2 hook and plugin directory packages (#915)
* fix(opencode): preserve v2 directory packages

Stage the reviewed issue 913 source for isolated fork validation. Preserve
package directories and manifests without executing managed packages or
installing their dependencies, and include receipt and localized UI fixes.

Reviewed local source: 15ac88e2f266571b8a6dc8d777302db0f4dcae44
Exact source tree: 45ebabfe2b

Static checks and independent inspection completed. Go, integration,
dashboard, rendering, and hosted runtime validation remain pending.
This fork staging commit is not an upstream publication or merge.

* test(opencode): align schema and snapshot assertions

Check the version-aware binding schema and verify native unregister retains the unchanged managed snapshot.

* fix(opencode): resolve Windows npm native launchers

Probe recognized native Opencode executables without invoking a batch shell; exercise the default version probe on hosted Windows.

* test(opencode): normalize Windows fixture line endings

Avoid doubling carriage returns when Git checks out the npm fixture with CRLF.

* fix(opencode): keep an undetected OpenCode version from blocking other agents

Review of the v2 directory package change found that one failed
`opencode --version` probe failed the whole hooks plan, so a machine
without the binary (a second machine sharing the config, CI, a
devcontainer) could no longer sync hooks to any agent, and doctor and
the dashboard preview failed with it.

Hooks:
- A probe that fails or reports an unknown major now skips only the
  OpenCode bindings with one warning per target. Outputs already owned
  there are parked, so they are neither rewritten nor removed while the
  layout is unknown. The probe runs once per plan, list and import.
- A `package` binding on OpenCode v1 is skipped with a warning instead
  of failing the plan.
- The v2 dependency warning is reported once, and only when a package
  file is written, so a synced hook no longer keeps doctor yellow.
- `hooks import` reports an unreadable file such as a symlink as that
  package's problem; other packages and flat files stay importable.
  The package validation problem is reported once instead of twice.
- Directories under `plugins/` are listed as unmanaged only on v2.

Plugins:
- The inventory leaves `plugins/skillshare-NAME` registrations to
  hooks, so a synced v2 hook is not offered as an importable plugin.

Dashboard: the new warnings are translated in every locale, a
`hook NAME:` prefix no longer defeats translation, the native preview
translates its per-file error, and the YAML editor documents `package`.

Docs: the English pages and the built-in skill describe the behavior
above; the ja, ko, zh-Hans and zh-Hant pages now carry the same
content as the English ones. The runbook drives the CLI with mdproof
assertions instead of prose expectations.

Refs #913.

---------

Co-authored-by: Willie <egg358358@gmail.com>
2026-11-16 01:50:14 +08:00
Willie 1f0077d2db chore(mcp): guard the conformance script and close the gaps left by #905 (#911)
Follow-up to #905.

The conformance script builds and serves skillshare and runs npm ci, which
the repository only allows inside a container, but nothing stopped it from
running on the host. It now refuses outside a container, with the same check
scripts/hooks/git-e2e.sh uses. Nothing referred to the script either, so the
testing topic now says when and how to run it.

The translated mcp reference pages left out the example URI the English page
gives for a skill root. All four now carry it.

The streaming UTF-8 check in the package loader was only exercised with a
rune split one byte before the read boundary. Two fixtures in the existing
directory test add the 2+2 and 3+1 splits, so a mistake in the carried-over
bytes shows up as a wrong MIME type.
2026-11-15 00:37:15 +08:00
Rusty Shackleford e969ebcfb0 fix(mcp): close Skills conformance gaps with directory reads and resource metadata (#905)
* feat(mcp): support Skills directory reads and metadata listing

Expose accepted skill directories and SKILL.md metadata so MCP hosts can
browse served packages without guessing manifest filenames. Reuse catalog
selection and refresh boundaries, including empty directories and stable
pagination, and normalize strict frontmatter for YAML 1.2 Core readback.

Add regressions and a pinned official conformance runner, and update the
runbook and MCP references to describe the browsing contract.

Refs #895

* fix(mcp): preserve supporting file MIME types in directory listings

Record UTF-8 classification from the same bounded bytes used for package
digests so directory metadata agrees with resource reads for extensionless
and unknown-extension text. Reuse the existing bounded reader without
adding listing-time payload reads or changing package limits.

Share pagination with skills/list, trust the SDK's typed request boundary,
and repair the pagination regression to detect stale catalog snapshots.

Refs #895

* fix(mcp): stream supporting file metadata inspection

Buffering supporting files during catalog refresh allocated with asset size.
Restore one bounded stream for digest, size and UTF-8 classification, retaining
at most three incomplete rune bytes and consuming invalid text through EOF.
Preserve the existing same-file checks, package caps and SKILL.md byte reader.

Keep real package-load benchmarks and strengthen the existing MIME fixtures
for split runes and bytes after invalid sequences. Use one first-match sed
command for the conformance URL without a truncated pipeline under pipefail.

Scoped packages, focused race tests, vet, shell syntax/shellcheck and all 45
pinned official checks pass. Allocation stays near 48 KB for 64 KiB and 8 MiB
assets; multibyte UTF-8 validation still costs CPU versus digest-only loading.

* test(mcp): consolidate directory and frontmatter coverage

Check every directory continuation URI so a duplicated middle entry cannot
hide a missing resource. Keep the existing before/after-cursor refresh cases.

Merge nested-skill acceptance fixtures in the catalog owner, preserving all
skill-count, skip-warning and directory assertions across configured and
targetless cases. Move directory capability coverage to initialization and
merge scalar, alias and rejection cases under the existing parser test.

Give cached-page benchmark skills names that match their URI directories.
Narrow tests, benchmark smoke and vet pass; duplicate-result and stale-catalog
mutations both fail, then restored source passes. Production is unchanged.
2026-11-15 00:21:35 +08:00
Willie be8f8b2874 fix(hub): read the same tags for drafts and the index (#909)
The index and the draft builder read a skill's tags with different readers,
so the same SKILL.md could publish different tags depending on the path:

- The index accepted only the comma form (tags: git, workflow). A YAML list
  became "[git" and "review]", and a block list produced nothing.
- Drafts accepted only a YAML list, so the documented comma form produced no
  tags, and metadata.tags replaced the root field. After #901 the index
  ignores metadata, which made the two disagree there as well.

Both now call readSkillTags. It reads the root field only, in either form:
a list comes from the decoded frontmatter, and the comma form from the
existing line scan. The scan also unwraps a one-line list, so a skill whose
frontmatter YAML rejects on another line (for example an unquoted colon in
the description) no longer publishes bracketed fragments.

Tags under metadata are custom data and are not read by either path. The
skill format page says so in all five locales.
2026-11-14 23:52:45 +08:00
Willie c88fe15a17 refactor(frontmatter): simplify the root-line check and align custom-field docs (#908)
Follow-up to #901.

IsRootFrontmatterLine took the caller's trimmed line only to test for blank
and comment lines, and its callers disagreed on how they trimmed it
(ParseSkillName trimmed spaces and tabs on the left, the rest trimmed all
whitespace). The helper now trims the raw line itself, so there is one
definition of a blank line and one argument fewer to get wrong.

Two tests added in #901 covered behaviour that change did not touch. The
ParseFrontmatterFields row is removed, and the ParseFrontmatterMap test is
reduced to the one case that guards what the skill format page promises:
nested metadata mappings and lists are kept.

Docs: the skill format page now recommends metadata for custom fields, but
the "Version tracking" example in creating-skills still showed top-level
version and author. It now matches. The sentence saying sync preserves
existing last_edited values is shortened, because skillshare has no such
field and the wording suggested it does. Both edits are applied to en, ja,
ko, zh-Hans and zh-Hant.
2026-11-14 23:41:01 +08:00
Rusty Shackleford 990732b454 fix(frontmatter): isolate root identity and hub index tags (#901)
* fix(frontmatter): keep nested metadata out of skill identity

Preserve root indentation in the tolerant identity readers so nested name
and description fields cannot shadow a skill or create audit claims. Keep
each reader's locator, scalar conversion and streaming policies intact.

Prove existing structured provenance preservation and the namespace fix
with regressions, bounded fuzzing, and global/project CLI integration.
Document nested metadata and native-client compatibility in all format
page locales.

Fixes #894

* refactor(frontmatter): share root-line indentation checks

Use one root-line predicate for the tolerant identity readers and
ParseSkillName so their metadata namespace boundary has one owner.
Keep each caller's whitespace policy, locator, scalar handling and
early-stop behavior intact.

Retain the existing namespace regressions, fuzz invariant and CLI flow
coverage without adding a duplicate helper test harness.

* fix(hub): keep nested metadata out of index tags

Apply the shared root-line predicate to the hub index scalar tag reader so
custom metadata tags cannot replace root tags or create absent root tags.
Extend the existing namespace table rows to assert exact index tags while
preserving the separate structured-list metadata precedence policy.
2026-11-14 23:18:15 +08:00
Willie 58ccb2bcc5 fix(uninstall): keep repos installed into a tracked repo's legacy group (#900)
Uninstalling the _team checkout matched every tracked entry grouped as
"team", so a repo installed with --into team (team/_b) lost its record
and audit acceptances too. Since #880 keeps declarations that have no
checkout on this machine, reconcile could no longer restore that record,
and the next push removed it on every machine; only trash restore of
_team brought it back.

#880 already excluded _-prefixed repo directories from the legacy match
for unlink and declaration-only removal. Move that guard into
inLegacyRepoGroup so checkout uninstall and its trash copy follow the
same rule.

Fixes #898
2026-11-14 14:25:00 +08:00
3217d44881 fix(reconcile): keep tracked repo declarations this machine has not cloned (#880)
* test(install): reproduce loss of pulled tracked records

Exercise the two-machine push and pull sequence and reconciliation states so
missing tracked clones cannot silently erase shared install intent.

* fix(reconcile): preserve tracked records missing local clones

Pull transfers tracked repository metadata but not ignored clones. Keep the
missing records during global reconciliation so another install and push cannot
erase shared intent or accepted audit findings on every machine.

* test(reconcile): isolate tracked record fixtures

Use the shared sandbox fixture for both reconciliation tests so their home
and configuration paths remain isolated from the invoking environment.

* fix(reconcile): retain all sourced tracked declarations

Legacy tracked members and unfollowed source links can hide declarations
from the local walk. Preserve sourced tracked records regardless of their
path so unrelated installs cannot erase them or their audit acceptances.

Document restoring a missing checkout before uninstalling it and cover
migration, link collisions, aliases, and explicit removal.

* test(reconcile): drop fixtures existing tests already pin

The tracked-record unit test also seeded a missing plain record and a
live plain record. TestReconcileGlobalSkills_PrunesStaleEntries already
pins both outcomes: a sourced plain record whose folder is gone is
pruned, and one on disk survives. The tracked record with no source
still catches a fix that keeps everything or skips the prune.

The two-machine test passed HOME and SKILLSHARE_CONFIG to RunCLIEnv,
which already sets both from the sandbox for each child process.
Separate homes, XDG paths, client overrides and git config remain.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test(reconcile): retain tracked entry for unavailable link

Expect sourced tracked declarations to survive reconciliation with either
source-link policy, so the test matches the cross-machine retention contract.

* fix(config): prune legacy tracked members

Retain only sourced tracked repository declarations during reconciliation.
Migrated member records are not restorable as repositories and retaining them
would leave permanent misleading install requests and audit acceptances.

* fix(reconcile): preserve tracked intent across followed links

A source link can exist before a tracked repository declaration is pulled. Preserve that declaration when the linked folder is not a checkout so later reconciliation does not discard its record and audit acceptances.

* test(reconcile): cover grouped followed declarations

A followed group can contain the path of a pulled legacy repository declaration, which must retain its tracked intent while its key migrates.

* fix(test): scope collision fixture runner correctly

Keep the isolated CLI runner in the collision subtest that uses it so the integration test package can compile.

* docs(manifest): clarify tracked checkout recovery

A retained declaration cannot restore a checkout while another folder or source link occupies its destination, and restoration requires a reachable source. State those prerequisites in each manifest translation.

* fix(config): prune members under recorded repositories

A migrated underscore-named member can otherwise survive a missing checkout as a repository declaration. Check recorded ancestor paths while preserving ambiguous grouped declarations and unrelated aliases.

* docs(manifest): restore fresh-clone project guidance

Explain that project declarations without local install records remain available so teammates can install them after cloning.

* refactor(reconcile): drop the tracked-intent walker policy

Restore baseline normalization for present paths while retaining absent
repository declarations through the shared metadata predicate. Isolate the
reconcile fixtures without building a CLI and share native-client overrides
without changing the original two-machine record and audit regression.

Validation: gofmt, git diff --check and context-router check only.
Runtime, builds and tests were not executed under the source-only gate.

* fix: forget linked checkout records and declarations without checkouts

Extend the metadata and uninstall owners to remove a declaration without
reading or moving its occupant, contacting its source, or changing trash
and gitignore. A failed metadata save makes every metadata-only result fail.

Clean up global unlink records in CLI and API, preserve existing-path name
precedence, report record-only results truthfully, and list declarations in
the dashboard API with noCheckout. Keep project removal behavior unchanged.
Remove the obsolete grouped-link walker-policy assertion while preserving
grouped absent declarations and the original record/audit A/B regression.

Validation: source-only five-pass review, gofmt, git diff --check and the
context-router check. No runtime, builds, tests or frontend tooling ran.
Stop at commit 15 for the mandatory Luna checkpoint before UI and docs work.

* fix(server): reload metadata before removing declarations

Preserve CLI records and audit acceptances written while the dashboard is open. Resolve new declarations from disk and refuse stale-cache saves after a reload failure, retaining unlink success when its link already moved.

* feat(dashboard): remove a tracked repository record without a checkout

Show declaration-only rows without checkout actions or status promises. Reuse Uninstall with record-specific confirmation and feedback, keep real-checkout force handling, and refresh the missing-entry list after repository removal or unlink.

* docs: explain kept declarations and how to remove them

Document direct global declaration removal and optional restore in English and all four translations, including occupied paths and normalization after reconcile. Keep built-in guidance accurate and use the existing API removal result for dashboard success feedback after concurrent CLI changes.

* fix(dashboard): confirm restored checkouts after forceable refusals

A tracked declaration can be restored through the CLI while its dashboard
row still says there is no checkout. Treat a forceable uninstall refusal
as evidence of the current checkout and promote the existing confirmation
to its dirty or unreadable-status warning before allowing explicit force.

Author stale-row regressions for both forceable codes before the source
repair, and retain nonforceable failure and actual-result success feedback.
Validation: five serial source review passes, git diff --check, and the
context-router check. Tests, builds, runtime and frontend tooling were not
run under the source-only gate.

* test(server): persist overview declarations before routed request

The overview fixture added declarations only to the in-memory skills store,
which the real request's config reload replaced with on-disk metadata. The
hosted unit, offline, and Nix jobs consequently returned no tracked rows.

Save the complete fixture through MetadataStore before serving the request
and fail setup on a save error. Keep the middleware, assertions, and
production behavior unchanged.

* fix(server): prefer exact declarations before checkout shorthand

Removing a missing _team declaration from the dashboard could resolve the
different org/_team checkout by basename and trash it instead.

Reload metadata before declaration lookup, give exact declarations priority,
and preserve checkout resolution before non-exact declaration fallback.
Add routed absent/occupied regressions and checkout precedence guards, and
persist existing checkout fixtures so the metadata reload keeps their state.

Validation: gofmt, git diff --check, and context-router check passed.
Tests were authored but not run under the source-only fleet restriction.

* fix: preserve removal intent and report partial failures

Bind dashboard record-only removal to its canonical declaration and refuse
restored checkouts before checkout resolution or force. Refresh the row and
require another confirmation before moving a restored checkout to trash.

Keep unresolved uninstall inputs and their reasons in JSON failed results,
including all-invalid requests, partial successes and dry runs. Report unlink
metadata cleanup failures as partial results, refresh completed unlinks and
show localized warnings. Keep CLI and dashboard guidance aligned.

Regression tests were authored before source changes and remain unrun under
the source-only task restriction. Static formatting, diff and router checks
passed; no build, product runtime, frontend tooling or test suite was run.

* fix(unlink): log partial metadata cleanup failures

Write the unlink result after global metadata cleanup so load and save
failures retain their diagnostic and report partial success. Preserve the
successful link removal, warning, target and exit policy, and the existing
link, project unlink and native removal failure behavior.

Add the corrupt-metadata regression before the fix; leave it unrun under
the source-only authorization. Formatting, diff and context-router checks
pass; runtime verification remains with the coordinator.

* fix(unlink): forget records below the removed link name

Reconcile can normalize a followed link's record to untracked, and linked
collections can retain nested repository declarations. Clean the removed
namespace through the existing metadata path owner in both unlink callers,
including audit acceptances, while keeping declaration uninstall strict.

Author CLI and route regressions before the source fix for normalized and
nested records, sibling preservation, and subsequent bare install. Tests
remain unrun under the source-only task restriction; static checks only.

* fix(uninstall): refuse a restored checkout at execution

A declaration selected while absent can become a checkout before removal
runs. Recheck the resolved source-relative name in the shared metadata-only
execution owner and refuse before touching metadata, regardless of force.

Reuse the existing dashboard conflict response and refresh/reselection flow
for this refusal, preserving the earlier record-only intent guard. CLI
results already carry shared execution failures through text and JSON.

Author restoration, force, absent/occupied control and client boundary
regressions before production edits. Correct the unlink fixture name and
response-key assertion found in serial static review. All tests remain
unrun under the source-only restriction; formatting/diff/router checks pass.

* refactor(server): consolidate repository uninstall result handling

Resolve a declaration or checkout to one item and canonical name before
using the existing uninstall, conflict, log and success response tail.
Keep metadata-only removal unforced, preserve its error and cache recovery
behavior, and report movedToTrash from the selected item.

Preserve exact declaration precedence, record-only checkout refusals,
ordinary checkout errors and project behavior without changing tests.
Five serial source review passes, gofmt, diff checks and the context-router
check completed. Tests and builds remain unrun under the source-only brief.

* fix(unlink): preserve independent records and trash undo metadata

Limit unlink's legacy group matching to members so a retained declaration
such as team/_repo survives unlinking _team. Share snapshot and removal
matching while keeping generic uninstall's legacy contract unchanged.

Save the exact removed records, audit acceptances and target overrides
through uninstall's existing trash sidecar owner before pruning metadata.
Keep metadata on sidecar failure and preserve completed-unlink warnings,
fresh reloads, cache refresh and project-mode behavior.

Author owner, CLI and API regressions for namespace collisions, undo,
partial failures and project/empty controls. Source-only verification:
gofmt, diff check and context-router check passed; runtime tests, builds,
frontend checks and final make check remain unrun for root-owned gates.

* fix(install): preserve independent tracked declarations

Limit declaration-only cleanup to the precise namespace and legacy members already used by unlink. Keep independently installed prefixed repositories, their audit acceptances, and overrides when forgetting a missing checkout.

* fix(cli): explain missing tracked repository recovery by mode

Offer global record removal when restoring an unreachable source is unwanted. Keep project recovery scoped to install -p, share the hint with doctor, and align update documentation in every existing locale.

* refactor(uninstall): share declaration and unlink operations

Keep declaration resolution and unlink metadata ordering in shared owners so CLI and API preserve the same scope and recovery behavior. Consolidate domain regressions while retaining caller translation coverage.

Distinguish record-only removals in API and CLI operation logs and CLI JSON, including mixed batches and dry runs. Document outcome fields, warning semantics, and current API error codes across existing locales.

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Willie <egg358358@gmail.com>
2026-11-14 13:11:27 +08:00
Willie cfe235a84d fix(install): report reinstalls from the same repo as skipped (#887)
* fix(install): report reinstalls from the same repo as skipped

checkExistingConflict read only the legacy .skillshare-meta.json
sidecar, which installs no longer write, so reinstalling a skill from
the same repo was reported as "already exists" and counted as failed,
contrary to the documented skip. It now reads the skill's record in
.metadata.json, falling back to the sidecar, and skips only the same
skill of that repo: another skill installed under the same name stays
a conflict instead of being reported as installed.

The dashboard logged an existing agent as installed and a same-repo
skill as failed. Batch installs now return those as skipped and log
them as skipped_skills, passing --into so a reinstall into a group
folder finds its record; the install dialog warns that they are
already installed. A same-repo tracked reinstall returns 409 and is
logged as skipped instead of a 500 logged as failed.

ss search installs, batch and single, log the names they skip, and
the Herdr panel reports skipped names like failed ones. The install
docs and their translations say that only the same skill is skipped.

* fix(install): install requested agents when the only skill is skipped

Reinstalling a single-skill repo with `-s alpha -a helper` returned as
soon as the skill was skipped as already installed, so the requested
agent was never installed while the command reported success. Both
single-skill branches now continue to the agent phase, as the
multi-skill path already does.
2026-11-14 00:03:17 +08:00
Willie 51306f49fa fix(install): log skipped skills and agents (#886)
After #884, an install without --force keeps existing skills and agents
and names them in the --json skipped array, but the operation log
recorded only installed and failed names, so `ss log` could not show
what an install kept.

Record the kept names as skipped_skills for CLI installs and for the
dashboard's install-from-config action, and list them in `ss log`.
The config install starts with an empty list, so the dashboard log
shows no "null" when nothing was kept.
2026-11-13 22:34:40 +08:00
Rusty Shackleford 4596c1e413 fix(install): require explicit force for JSON overwrites (#884)
* fix(install): require explicit force for JSON overwrites

Machine-readable output must not grant destructive consent. Preserve existing skills and agents unless force is explicit, and report existing skip decisions in the JSON result while retaining selection and audit behavior.

Cover edited copies in global and project modes, and update help, automation guidance, and translated command documentation.

Refs: #877

* fix(install): preserve previews and locked JSON copies

Read-only discovery must work before replacement consent, including existing
single skills and cross-path installs. Keep overwrite checks on write paths.

Require existing update or force intent to advance installed project copies
to a changed lock pin. Preserve ordinary CLI and dashboard lock following,
while JSON alone leaves local edits and incoming pins unchanged.

Cover existing-destination previews, edited locked copies and dashboard lock
following. Clarify translated docs and make the consent runbook step runnable.

Refs: #877

* fix(install): resolve installed names during previews

JSON previews must resolve recorded sources without requiring overwrite consent. Permit metadata lookup for dry runs and restore the original unforced preview regression in global and project modes.

* test(install): remove duplicate JSON consent runbook case

The global local-skill integration case already covers JSON overwrite refusal and explicit forced replacement. Remove the duplicate runbook step and restore its 18-step numbering while retaining the existing purity setup and assertion corrections.
2026-11-13 21:31:41 +08:00
Willie 08ecb7afad fix(audit): report audit-rules load errors and block installs whose audit cannot run (#882)
* fix(audit): report audit-rules load errors in global scans

The global scan entry points (ScanSkill, ScanSkillWithFollow,
ScanSkillFiltered, ScanFile) dropped the error from Rules() and scanned
with no pattern rules. One bad regex in the global audit-rules.yaml
turned every pattern check off with no warning, so audits reported
fewer findings than they should have (#879). Project mode already
returned the error.

Return the error from the global path as well, and name the failing
file in it: parse errors from loadUserRules and compile errors from
Rules() and RulesWithProject now carry the path. When a project scan
fails to compile, the global file is blamed only if it fails on its
own, so a project file that disables a broken global rule still loads.
The cached markdown split now takes the rules its caller already
loaded instead of loading them again and dropping the error.

* fix(install): block installs whose audit cannot run

Once global scans return rule-load errors, install treated the error as
a warning and installed the skill with no audit at all. update already
fails closed on a scan error, and --force there accepts findings, not
an install nobody could check.

A scan error now blocks skill, agent and tracked-repo installs and
removes what was installed. --force does not override it;
--skip-audit still bypasses scanning. The error wraps the new
install.ErrAuditScanFailed rather than audit.ErrBlocked, so install
output and the dashboard do not offer a --force retry that cannot help.

upgrade reinstalls the built-in skill with --force, which deleted the
old copy before the audit ran. A typo in audit-rules.yaml would have
left users without the skill, so upgrade now stages the download
through the update path and keeps the installed copy when the audit
fails.

update --all counted the new error as a silent skip. It is now listed
under "Blocked by security audit" with its reason and makes the run
exit non-zero. The install warning summary that counted "audit scan
error" warnings is removed, since nothing emits them any more.

Fixes #879
2026-11-13 15:14:05 +08:00
Willie 40697e03f3 fix(oplog): keep source credentials out of the operation log (#881)
An install source typed as a URL can carry a token in its userinfo or
query, and operations.log stored it as typed, so anyone reading the log
or the dashboard's log view could see the credential (#878).

Redact in oplog.Write, which every CLI and dashboard writer goes through:
the source, oldSource and newSource args lose their userinfo, query and
fragment, and the message loses the same parts where an error echoes
them. The redaction reuses update's redactRepoURL, moved to
install.RedactRepoURL. It now also covers git+https:// and ssh://
passwords, and a source typed without a scheme such as
user:token@host/owner/repo, which install clones over HTTPS.

logInstallOp logged args[0] as the source, which is a flag when one comes
first (install --into team <url> logged "--into"), so the message was
never redacted. It now logs the parsed source.

Fixes #878
2026-11-13 14:50:31 +08:00
Willie 0939eb9ad9 fix(plugin): verify native support only for changes that run a command (#873)
* fix(plugin): verify native support only for changes that run a command

prepareChange probed native command support before applying the
intentional skip rules. When the probe failed under update --all, the
change was marked as a failed target, so the later skip never applied.
An imported Grok plugin always hit this: Grok updates require native
trust, so every update --all exited nonzero even though Grok plugins are
documented as updated natively. #853 patched the same ordering for
imported OMP plugins only.

Run the probe after the skip rules instead, so a skipped change is never
verified, and drop the OMP-specific exemption it made redundant. Host
errors and changes that do run a native command still fail as before.

* docs(plugin): align OMP removal wording and runbook pass criteria

The translated plugin references said OMP removal requires a
scope-matching native registration. English, and the code, only require
that a registration match when one is present; a retry after native
uninstall proceeds from the retained cache. The runbook's Pass Criteria
also did not cover the two scenarios #853 added as steps 4 and 5.
2026-11-12 14:09:10 +08:00
Rusty ShacklefordandWillie caa107b880 fix(plugin): refresh stale OMP catalog and report blocked OMP updates (#853)
* fix(plugin): recover OMP stale catalog installations

Refresh owned marketplace catalogs before reuse and validate the native cache identity when removing an installation. Preserve that identity through interrupted cleanup and report blocked batch updates as failures so retries remain safe and visible.

* fix(plugin): preserve reviewed versions during OMP recovery

Keep native removal identity separate from reviewed receipts so interrupted cleanup can retry safely. Expose isolated failures in previews and dry-run exit status while allowing healthy targets to proceed.

* fix(plugin): resume interrupted OMP update cleanup

Retry matching pending removals before reinstalling after registry-first interruption, and clarify isolated failure reporting. Strengthen recovery fixtures and restore accurate installation guidance.

* test(plugin): model OMP cached catalog installs

Use the cached marketplace entry for native install version and source so the stale catalog regression exercises the real version-to-cache relationship.

* test(plugin): align OMP batch expectations with safety checks

Require catalog refresh before reinstalling from an owned marketplace and assert isolated failures when native OMP ownership cannot be verified, while healthy targets update.

* fix(plugin): skip native upgrade checks for OMP imports

Imported plugins have no reviewed source to upgrade, so requiring native upgrade capability makes a safe batch skip fail unnecessarily. Cover that case and OMP dry-run failures, and align translations and native verification runbook with the failure and removal behavior.

---------

Co-authored-by: Willie <egg358358@gmail.com>
2026-11-12 13:45:55 +08:00
Willie 9dbcd925d8 feat(audit): honor findings accepted with --force via --respect-accepted (#872)
* feat(audit): honor findings accepted with --force via --respect-accepted

install and update --force record the findings a user accepted in
audit_accepted, but audit ignored that record. A skill installed with
--force therefore kept failing `skillshare audit` in CI with no way to
let the accepted findings pass (#869).

audit --respect-accepted marks those findings as acknowledged. They stay
in every output (JSON acknowledged: true, a text/TUI note, "(accepted
earlier)" in Markdown, SARIF external suppressions), but they are counted
in summary.accepted instead of their severity and no longer fail the
audit. Without the flag nothing changes, so existing CI gates keep their
behavior. A changed file or snippet produces a new key and blocks again.

Tracked repositories record acceptances at the repo root with paths
relative to it, so ApplyAcceptedFindings walks up from each skill to
match them. The install, update and edit gates use the same function and
now honor repo-wide acceptances for skills inside tracked repos too.

The dashboard always honors accepted findings: the Audit page mutes them
and counts them apart, the skill Audit tab collapses them, and the
Dashboard attention card no longer lists a skill whose only blocking
findings were accepted. The CLI JSON summary and the server summary share
Result.CountOpenBySeverity so both count accepted findings the same way.
The oplog records the accepted count and the log views show it.

* fix(audit): match acceptances when a single file is scanned

A single-file scan names its findings by the file's base name, but the
ancestor lookup used the file's whole path as the prefix, so it hashed
SKILL.md/SKILL.md instead of the recorded SKILL.md. Auditing a skill file
directly with --respect-accepted, and auditing agents inside a tracked
repository in the CLI or the dashboard, still failed on unchanged
accepted findings. For a file target the prefix is now the file's own
directory.
2026-11-12 11:15:44 +08:00
Willie 65f3c8d1b9 fix(ui): show health checks as they finish instead of after the slowest (#870)
* fix(ui): show health checks as they finish instead of after the slowest

The Doctor page waited for the whole `doctor --json` run before showing
anything. One slow check, such as plugin checks waiting on a native CLI
that does not answer, kept the page on a skeleton for 30 seconds, and the
server's 30-second write timeout could cut the response off entirely.

`doctor --json` now prints each check as a {"check": ...} line when the
internal _SKILLSHARE_DOCTOR_STREAM variable is set, before its usual
result. The new GET /api/doctor/stream relays those lines as Server-Sent
Events and sends the full result as `done`; it has no write deadline and
stops doctor when the client leaves. The page shows checks as they
arrive and replaces them with the final result.

* fix(ui): show the server's reason when a stream fails

The audit, diff and doctor streams send an `error` event with the reason,
but the shared EventSource helper treated it like a dropped connection
and showed only a generic message. It now uses the reason the server
sent, and keeps the generic message for a real disconnect.

* fix(doctor): skip Git reads in the hooks check when no hooks are configured

The hooks check loaded the full hooks inventory, which also scans
unmanaged native and Git hooks twice: about 12 git calls even with no
hooks configured, none of which doctor reports. Doctor now reads the
hooks section first and previews a sync only when there is something to
check.
2026-11-12 10:40:15 +08:00
Willie 9f43310a3d fix(update): repair skill sources recorded before the per-skill source fix (#866)
Before #863, `install <repo> -s <skill>` recorded "<repo>/<path>" as the
skill's source. GitLab, Bitbucket and file:// sources parse that string as
another repository, so every path that parsed the recorded source failed
for those skills: `update` of a single skill, `install <name> --update`,
the dashboard update, and the local-changes diff. #863 only changed what
new installs record.

The record still holds the repository and the path, so RecordedSource
rebuilds the source from them when the written string parses as a
different repository, and the reinstall writes the record back in the
form that parses.

In project mode the config entry that declared the old string follows the
rewritten record, because teammates install from the config. An entry a
teammate changed stays as it is, and so does a record a batch update only
respelled. The config is reloaded first: `update --prune` saves it through
another copy, and saving the stale one brought pruned skills back.

`hub draft` and `hub index` published the old string as it was written;
they now publish the source that installs the skill.

Closes #862
2026-11-11 23:34:43 +08:00
Willie c040917bc7 fix(security): enforce credential and filesystem trust boundaries (#861)
Bind provider credentials to canonical or explicitly configured HTTPS origins and preserve that policy across remote consumers. Reject cross-site credentialed reads, linked fallback directories, and discovered-agent writes outside the source; audit staged content before rooted atomic publication.

Add boundary regressions, private Git update coverage, stable cross-platform fixtures, and aligned documentation. Verified the integrated main baseline with all executed CI checks passing.
2026-11-11 23:17:54 +08:00
Willie 3872cb5525 fix(install): record a per-skill source every host can install again (#863)
* fix(install): record a per-skill source every host can install again

`install <repo> -s <skill>` joined the repository and the skill path with
a single "/". GitLab and Bitbucket URLs have no fixed depth, so the parser
read the result as a longer repository path, and file:// URLs only split
on "//". The recorded string is what a project's config.yaml shares, so
`install -p` on a fresh clone tried to clone a repository that does not
exist.

Whole-repo file:// sources now join with "//", and GitLab or Bitbucket
sources typed without ".git" or a web URL marker join with ".git/". The
other hosts keep their form.

`install --update` still recognises a record in the old form, but only
when its recorded repository and path agree, because on these hosts the
same string can also name a repository of its own. The dashboard's
install dialog matched installed skills by rebuilding the old form, so it
now accepts each form the server writes.

Closes #859

* fix(ui): mark hub results installed under the new recorded source

A hub entry that names a repository and a skill was matched against the
installed skills by an exact source or a "<source>/" prefix. A GitLab or
Bitbucket repository typed without ".git" is now recorded as
"<source>.git/<skill>", so the search row and the preview kept offering
Install, and a second install hit the already-exists warning.

The matcher now accepts the ".git/" join as well. This is the same class
as the discovery list fixed in the previous commit: both rebuild the
recorded source in the dashboard. No other dashboard or Go code matches a
recorded source by prefix.
2026-11-11 23:14:39 +08:00
Willie 769087b882 fix(project): keep config.yaml true after reinstall, uninstall and fresh-clone installs (#860)
In project mode the committed config.yaml is what teammates install from, but
it could disagree with what a machine actually has (#858):

- `uninstall -p` and the dashboard's single-skill and tracked-repo removal
  left the skill's entry in config.yaml, so a later `install -p` brought it
  back. They now drop the entry and the lock pin through one shared
  config.ForgetProjectSkills, as the uninstall docs already promised.
- Installing a skill again from another source or branch (after an uninstall,
  or over the copy with --force) updated only the gitignored .metadata.json.
  config.yaml kept the old source and skills.lock.json ended with the old
  source and the new commit. Installs from a named source (CLI install,
  search, dashboard) now compare the records from before and after the
  command and let the config entry follow the skills it actually reinstalled.
  Install from config never rewrites an entry from local records, so a
  teammate's edit survives on a machine that has not updated yet.
- Reconcile removed every config entry that had no folder and no record. On a
  fresh clone nothing is installed yet, so `install <source> -p` deleted all
  the other declared skills. An entry now goes only when the skill was
  installed here and its folder is gone, when it was moved, or when a command
  removed it; `move` passes the old paths and `update --prune` cleans the
  config itself.
2026-11-11 22:42:03 +08:00
Willie 828ba93f27 fix(git): keep a caller's GIT_DIR out of skillshare's git commands (#854)
* fix(git): keep a caller's GIT_DIR out of skillshare's git commands

Git exports GIT_DIR, GIT_WORK_TREE, GIT_INDEX_FILE and related variables
to hooks that run in a linked worktree. Every git process skillshare
started inherited them, and those variables override the directory the
command runs in. Run from a post-checkout or post-merge hook, skillshare
therefore acted on the hook's repository instead of its own: install's
sparse clone turned on sparse checkout there (a later merge removed every
file outside the skill's path), the remote tree check's `git init --bare`
made it a bare repository, install's reconcile recorded its origin and
branch, and push staged and committed into it.

internal/gitenv builds git commands without the variables that
`git rev-parse --local-env-vars` lists for choosing a repository, work
tree, index, object store or config file. GIT_CONFIG_COUNT and
GIT_CONFIG_PARAMETERS stay, because users pass credentials through them
and install's token auth appends to the existing count. Every git call in
cmd/skillshare, internal/install, internal/git, internal/check and
internal/config now goes through it, and plugin runCommand starts native
CLIs from the same environment because they run git as well. The hooks
service keeps its own stricter filter.

Closes #852

* test(gitenv): fail on new git commands that bypass gitenv

Each git call site converted for #852 is safe only because it goes
through gitenv.Command, and about half have no test of their own. A
future exec.Command("git", ...) would silently bring the bug back, so a
syntactic guard now rejects one outside gitenv, hooks and testutil,
mirroring the sourcefs and sourcewalk ratchets.

The Environ test also checks a lowercase name, since Windows environment
names are case-insensitive, and compares against literal keys so that
removing one from repoVars fails it. The wiki records the rule.
2026-11-11 21:08:47 +08:00
Willie adf6c9ecb3 feat(updates): allow opting out of automatic version checks (#847)
Add a global automatic-update preference with an environment override, CLI controls and dashboard settings. Suppress automatic release checks and cached notices while retaining explicit checks and upgrades.

Recheck policy before publishing in-flight results, preserve structured JSON errors, and reject incomplete manual checks. Include schema, completion, localization, documentation and regression coverage.

Closes #846.
2026-11-11 11:59:57 +08:00
Willie bce18d2e19 fix(audit): resolve a nested skill's references against its own root (#845)
Auditing a parent skill checked every Markdown file's bundled references
against the scan root. A nested skill's `references/ok.md` was looked up
under the parent and reported as a dangling link, and its `../SKILL.md`
stayed inside the parent and was not reported as an escape, so a parent
audit disagreed with a direct audit of the child.

Each Markdown file now records the nested skill that holds it, and its
references are resolved and contained within that skill.

doctor scans a nested skill on its own and again inside its parent, so
it repeated the child's findings. It now reports each finding once.

Closes #834
2026-11-11 00:48:16 +08:00
Willie 4556438575 fix(plugin): let OMP and Pi plugin updates through without uninstalling (#833) (#837)
* fix(plugin): let OMP and Pi plugin updates through without uninstalling (#833)

Every OMP update was blocked, because OMP's native upgrade replaces cache
folders that projects outside its inventory may still use. The only way to
move an OMP binding to a new commit was to uninstall first. An update now
runs the same pair a source change already uses: the scoped removal
adapter (cache kept), then an install under a fresh marketplace and cache
identity. A refresh always takes a new identity, because OMP installs from
the catalog it cached when the marketplace was added, and that catalog
still names the old version. If the removal cannot be verified, the
update is blocked; update --all still skips it with the reason.

Pi was blocked whenever the catalog plugin name differed from its
package.json name. That happens after an upstream npm rename, for example
when npm refused the original name. Pi installs by path, and Skillshare
keys the binding by the catalog name, so the npm name never reaches
either, and the check only stopped working installs.

Updating one package across several Agents stopped every Agent as soon as
one was blocked. Each failing Agent is now reported with its reason while
the others update, and the command exits non-zero. A single --target, or
a package with no other Agent to update, stays blocked as before, so a
dry-run still fails when nothing would update.

* fix(plugin): let one update --all refresh several OMP packages

Every OMP removal proves itself against a fingerprint of the whole native
registry and lock. In a single update --all plan, the first package's
removal and reinstall rewrite those shared files. The second package's
removal therefore saw "OMP native files changed", so only one OMP
package could update per run. The same applied to a sync that replaced
the sources of several OMP packages.

A removal now also fingerprints only its own installation: its registry
entry, its lock entry, its package manifest and its runtime link. After
an earlier change in the same apply has written to that OMP target, the
removal re-reads the files and runs every ownership check again. It goes
ahead only if that own fingerprint still matches the preview. A change
outside Skillshare to the package's own entries still blocks it. Without
an earlier write in the apply, the full-file check is unchanged.

The fake OMP in the tests now merges installs into the registry and lock
and lists them in a stable order, as the real one keeps other plugins.
2026-11-10 21:46:42 +08:00
Willie 816938c09c fix(cli): report why uninstall and install fail under --json (#835)
Scripts and the herdr panel could not tell why an uninstall or install
failed: uninstall --json had no error field, project mode ignored --json
(printing text and still prompting), and a multi-skill or config install
in which every skill was blocked exited 0.

- uninstall --json (skills and agents, global and -p) sets "error" with
  each failed name and reason; skills exit non-zero when all fail,
  agents when any fails. Project mode now emits pure JSON, including on
  the first run that initializes the project.
- uninstall -p agents parses its flags instead of treating --help or an
  unknown flag as an agent name, and uninstall agents --json is logged.
- install exits non-zero, naming each skill and reason, when every skill
  from a source or from config fails; installed or skipped skills make it
  a partial success that still exits 0. Config installs show each
  failure's reason, and the dashboard logs an all-failed run as an error.
- herdr asks for JSON in project scope and shows the CLI's error.
2026-11-10 20:24:44 +08:00
Willie c74a35dd4b fix(audit): text-scan hidden-directory skills in tracked repo gates (#827)
Tracked install and update audit the whole repository root, and the
scanner treated every file below a hidden directory as header-only. Skills
that discovery finds in hidden directories, such as .curated/<skill> and,
since #801, a repository-root .agents/skills/<skill>, were therefore synced
to every target without a text scan.

The hidden-directory boundary is now measured inside the skill that holds
a file, so those skills get the text scan they already get when audited one
by one, while target copies such as .claude/skills and hidden folders
inside a skill stay header-only.

Release-As: 0.30.3
2026-11-10 16:21:26 +08:00
Rusty Shackleford 71ff2d9e14 fix(discovery): find repository-root .agents/skills (#801)
A repository-root .agents/skills now counts as a skill anchor only when the repository has no other SKILL.md outside .git and target directories, so tracked checkouts that author skills there are discovered, while repos that commit .agents/skills copies keep discovering only their authored skills.

Fixes #800
2026-11-10 16:04:20 +08:00
Willie cafd99e595 feat(mcp): add Hermes Agent sync and import target (#826)
Share portable MCP definitions with Hermes through its native YAML configuration while preserving unrelated settings and managed-entry safeguards.

Support explicit Hermes homes and native OAuth options without copying credential state, and expose the target consistently in the dashboard, schemas, completions and documentation.

Refs #825
2026-11-10 15:19:11 +08:00
Willie bb514c9432 fix(git): skip gitignored tracked repos in the root-scope nested repo check (#821)
With git_root: root, commit and push aborted on every tracked skill repo
because NestedRepos flagged any directory with its own .git. Tracked repos
are listed in the managed .gitignore block, so git add -A never stages
them, and the suggested remedy (renaming .git to .git.disabled) breaks
skillshare update for those repos.

NestedRepos now asks git check-ignore about each nested repo and leaves
out the ones git ignores and does not track. A repo already committed as
a gitlink is still reported, since staging would update it. Paths are
checked one at a time because git refuses a whole batch when one path
lies inside a gitlink, and any git error keeps the warning.

Closes #820
2026-11-10 01:00:03 +08:00
Willie 037c7943cf fix(audit): report credential paths in Python comments and docstrings as HIGH (#819)
Credential-access rules match whole lines, so a Python docstring that
named ~/.aws/credentials in a `cat` example counted as a CRITICAL read and
blocked the mcp-sentinel skill at the default threshold (#817).

In .py files, a built-in CRITICAL credential-access match on a comment-only
line, or inside a string that is a whole statement such as a docstring, is
now HIGH: it still shows, and still blocks under a HIGH threshold. A small
tokenizer decides this and keeps CRITICAL wherever it cannot be sure the
line never runs: argument, assigned and bracketed strings, f-strings and
their replacement fields, code before a trailing comment, files that read
__doc__ or use bare CR line breaks, and explicit severity overrides.

A lowered finding carries severityContext "python-inert" (JSON, SARIF,
`audit rules --format json`), and the dashboard, CLI text output, both
audit TUIs, install/update warnings and the markdown report say it was
lowered from CRITICAL, so it is not mistaken for a changed rule.

Closes #817
2026-11-09 23:30:37 +08:00
Willie 8632fe685f fix(server): restart the dashboard through Homebrew's link after an upgrade (#816)
On a Homebrew install the dashboard runs from a versioned keg that `brew upgrade` removes, so the restart after "Update now" exec'd a deleted path. It now restarts through <prefix>/bin/skillshare, which Homebrew keeps linked to the installed version.

Fixes #815
2026-11-09 21:49:52 +08:00
Willie 2bc791ed19 perf(plugin): check every plugin at once and skip sources that did not move (#812)
* perf(plugin): check every plugin at once and skip sources that did not move

The Plugins page's update check read every Agent CLI, cloned and hashed
every source, and asked npm about every package one after another, so it
grew with each plugin added. A single Agent CLI that stopped answering
held the whole page at "Checking for updates" with no way to tell which
plugin was stuck.

The check now asks every Agent and checks every plugin at the same time.
Plugins share one read of each Agent and each source, so an Agent with
many plugins is still asked once, and a plugin waits only on its own
Agents, including when proving its own leftover installs. Before downloading a remote source with no pinned ref, the
check asks the remote which commit its default branch points at with
`git ls-remote` and reports no update when that is the installed commit.
A check of three unchanged GitHub plugins on two Agents went from 4.3s
with three clones to 0.6s with none.

The dashboard reads the check from a new stream, GET
/api/plugins/check/stream, that sends each plugin as soon as it is
checked. The page marks rows as they arrive and counts progress in the
header. The stream stops waiting after 60 seconds: a plugin bound to an
Agent that has not answered by then shows "Check failed" with the
reason, and plugins on other Agents are unaffected. There is no limit on
how many plugins run together, because plugins waiting on a hung Agent
would otherwise hold every slot and starve the rest.

A cancelled command did not return when the Agent CLI was a shim that
starts the real client, such as an npm, mise or volta wrapper: killing
the shim left the child holding its output, so the command waited until
the child exited and no timeout took effect. Commands now stop waiting
for output two seconds after they are cancelled; a command that exits on
its own while a helper keeps its output open still counts as answered.

* fix(ui): offer every update the Plugins page shows in the update dialog

The header counted updates from the per-plugin marks, but the update
dialog offered only the targets in the last completed check's plan. Two
cases left the header saying Update N with nothing to tick in the dialog:
a first check that broke off after finding an update never set the plan,
and a recheck that could not reach an Agent kept that Agent's earlier
update while the plan said blocked.

The dialog's plan is now derived from each plugin's last answer together
with the kept updates, so a kept update stands in for the blocked or
missing answer and both always agree. A broken-off check also keeps the
answers it did receive.
2026-11-09 15:06:55 +08:00
Willie e9dc4d0360 feat(server)!: answer only at localhost, IP addresses and trusted origins (#811)
DNS rebinding lets a page point its own domain at the user's machine,
which makes its requests same-origin to the dashboard, so the CSRF
guard cannot refuse them. Only the MCP, Hooks and Plugins routes
required a localhost or IP Host, so such a page could still install,
sync and push. That rule also blocked those three pages for people who
run the dashboard behind their own reverse proxy (#809).

The dashboard now answers every route only at localhost, an IP address,
the --host name, and origins given with --trusted-origin or
SKILLSHARE_UI_TRUSTED_ORIGINS. Configuration stays outside config.yaml
so the dashboard cannot widen its own trust through its config API. A
trusted origin must match the Host with its scheme, and an Origin
header, when sent, must be that origin. A refused browser gets a page
with the exact setting to copy, the API gets it in the error, and the
server log names each refused Host once.

BREAKING CHANGE: a dashboard opened by a domain name, such as a reverse
proxy, a Tailscale or mDNS name, or a Docker service name, now returns
403 on every page until that address is trusted, for example with
`skillshare ui start --trusted-origin https://skills.example.com`.

Release-As: 0.30.0
2026-11-09 11:59:50 +08:00
Willie ec9b3e688c fix(server): refuse dashboard changes sent from other sites (#810)
A web page can submit a form or a no-cors fetch to the local dashboard
without a CORS preflight, and the handlers accept a text/plain or empty
body. Only the MCP, Hooks and Plugins routes checked where a request
came from, so any site the user visited could add targets, install,
sync, commit, push or upgrade, whenever the browser let the request
reach localhost.

Every POST, PUT, PATCH and DELETE now needs Fetch Metadata, when sent,
to say same-origin, and Origin, when sent, to match the Host. A client
that sends neither, such as curl, is unchanged, because a page cannot
make a browser omit both on a cross-origin request. The update stream
runs updates over GET, as EventSource only opens GET, so it gets the
same check.

The nginx example now forwards $http_host, which keeps the port, since
a proxy that drops it would now fail every change, not only MCP.
2026-11-09 11:08:43 +08:00