* fix(update): audit tracked candidates before publication
Prevent rejected tracked-update bytes from becoming readable through published target links by staging and auditing a private candidate before updating the active checkout. Publish the exact approved revision without a second upstream pull.
Preserve sparse state, audit metadata, followed checkouts, and compatible edits. Recover staged and unstaged changes after failed forced publication, including deletions, renames, and sparse exclusions. Cover the boundary deterministically and align error messages and localized documentation.
* fix(update): preserve local bytes across Git line-ending conversion
Overlay dirty worktree bytes after staging candidate patches so core.autocrlf does not silently rewrite local edits. Cover staged, unstaged, and cancelled worktree changes while retaining their index state.
Make permission-failure coverage explicitly POSIX-only and compare private checkout contents to their original bytes. Align red-team assertions with candidate rejection and register private temporary cleanup with the raw-write guard.
* test(install): normalize checkout after configuring line endings
Windows fixtures initially inherit core.autocrlf=true. Reset the isolated checkout after selecting the subtest setting so changing it to false does not turn untouched upstream files into local edits.
Use the existing YAML parser and serializer so folded descriptions, quoted text, and nested metadata survive repeated editor saves without accumulating escapes or losing structure. Keep legacy headers editable and render nested object fields as read-only JSON.
Add regression coverage for the MinerU workflow reported in #637 and document editor behavior. Verified UI tests, lint, build, make check, and an isolated browser save/reopen flow.
Keep local-edit statuses wide enough so they do not collide with update actions. Wrap update rows on narrow screens and show row actions on touch devices.\n\nReuse the shared dialog for mobile navigation, close it before blocked or successful keyboard navigation, and keep the language picker and tour within the visible mobile UI. Include locale labels, responsive layout documentation, and tour-step regression coverage.\n\nFixes #638.
* feat(check): show what changed in a skill since install
#622 made a hand-edited skill visible as modified and kept update from
overwriting it, but there was still no way to see the edits before
deciding to keep or discard them (#621).
LocalChanges compares the skill's files with the hashes recorded at
install, fetches the source at the recorded commit only when a recorded
file changed, and diffs each file against it, so the diff shows the
user's edits even after upstream moves on. A skill without a recorded
commit is refused rather than diffed against HEAD, which would show
upstream changes as the user's.
It backs `check <name> --diff`, GET /api/resources/{name}/local-changes,
and the dashboard: the Modified locally dialog on the skill page lists
the diffs, and the Updates tab links a modified skill to it. update's
skip hints now point at --diff. The route looks up the exact record and
refuses non-local names, so a nested or ../ path cannot borrow a
top-level skill's record or read outside the source. The file diff
helper moved to textdiff so the CLI and server share it.
* fix(install): diff local changes against the full installed commit
entry.Version holds the short SHA from `rev-parse --short`. A targeted
fetch needs the full SHA, so checkoutCommit fell back to fetching every
branch and tag with --unshallow: viewing one edited skill could download
the repository's whole history, and fail when the commit is only
reachable by its full SHA.
Use entry.Commit, the full SHA install records, when it is the same
commit as Version; a Commit left from an older install, or a legacy
record without one, falls back to Version.
* test(install): keep checkout line endings when editing in LocalChanges test
On Windows git checks the installed copy out with CRLF, so overwriting
SKILL.md with LF content made every line differ and the expected diff
failed. Edit the installed file in place, as an editor would, and compare
the diff without carriage returns.
* fix(update): honor accepted findings in dashboard tracked updates
The dashboard ignored findings accepted through the CLI, so a tracked repo
that `update --force` let through was blocked again on the next dashboard
update, and every dashboard Force Retry had to be repeated. Regular skill
updates in the dashboard already honored and recorded acceptances, so only
tracked repos differed.
Tracked-repo updates now always honor accepted findings and record the ones
an override lets through, for both callers. The opt-in option and the
AuditGate flag it fed are removed, since no caller turns them off.
Dashboard agent tracked repos passed an empty source dir, so a nested repo
kept acceptances in its group folder under a different key than the CLI.
They now pass the agents source, as the CLI does.
* fix(update): reload the agents store after agent repo refresh
Dashboard agent tracked repos now refresh metadata in the agents source,
but a metadata change reloaded the skills store, leaving the agents store
as it was until the next request. Reload the store that matches the source.
Every caller of update.TrackedRepo now passes a source dir, so the
empty-dir guard and its doc no longer describe a supported mode. The new
agent test reuses the tracked fixture instead of rebuilding it.
A pull with dozens of conflicting files meant clicking "Keep this
version" once per file. Two buttons now set every file to the same
side; individual files can still be switched afterwards.
Codex reviews often took four to six rounds because each round found
one more place where the same bug class was unhandled, and it re-raised
cases outside the supported workflows that only our internal wiki ruled
out. AGENTS.md now carries code review rules Codex reads: supported
scope, one finding per bug class with every location, and re-reviews
limited to new commits. The pull request workflow now fixes every
instance of a confirmed class at once and, after three rounds, explains
why findings keep coming instead of only stopping.
* feat(memory): add host-owned review and verification workflows
Share scoped memory tools between Dashboard and Memory, and route retrospective and Dreaming requests through the built-in skill. Keep agent execution in the host, require approval for reviewed edits, and preserve versioned CLI write safeguards.
Confirm scope before exposing prompts and include the resolved root in recent memory responses. Update provenance guidance and localized documentation, with regressions for unavailable scope and protected review prompts.
* fix(memory): distinguish trial copy controls for screen readers
Include the localized step title in each trial copy control accessible name while keeping the short visible label. Update the existing clipboard test selector for the distinct names; no new test or local test rerun was requested for this label-only fix.
* fix(memory): point connection hints to the trial tools
Replace stale verification-prompt directions in all eleven locales with the current Memory tools and Try across sessions labels. Interpolate the existing localized labels so future naming changes remain aligned.
No new tests or local test reruns for this copy-only correction, as requested by the maintainer.
* feat(ui): surface existing shared memory on the dashboard
Make shared notes visible through recent updates and direct links, with file timestamps and an introductory state for untouched starters. Keep the dashboard quiet until memory is configured, and preserve access to existing notes when starter files are missing.
* fix(memory): bound dashboard reads and remember preview expansion
Count vault entries from metadata and read at most the two starters and four recent candidates instead of hashing the full vault on dashboard load. Keep incomplete starter states accessible and refresh the overview after source and instruction changes. Remember preview expansion in browser storage and raise the collapsed height to 360px.
The move command page broke the locale builds because its relative
links and anchors only resolve against translated siblings; the
verification list now names the rule that avoids it.
* docs(wiki): cap Codex review rounds and require a state matrix before review
PR #513 reached 7+ Codex rounds because each round surfaced one more
state dimension (project mode, tracked installs, followed links, legacy
keys, lock pins) that the change had never listed. Every round costs the
maintainer a review cycle.
Require a state matrix and a consumer sweep before the first review,
separate findings worth code from unrealistic combinations, and stop at
three rounds unless a finding shows data loss or a broken supported
workflow.
* docs(proposals): add move installed skill proposal for #510
Moving an installed skill with mv strands its install record, and the
documented workaround (uninstall + install --into) re-downloads and
drops audit acceptances and lock pins. Propose one explicit move
operation shared by the CLI, the Web API and the dashboard.
Records the maintainer decisions: no auto-sync, followed links refused,
target filters warn only, whole-folder move in v1, --force limited to
name collisions.
* fix(proposals): cover nested skills, bare tracked checkouts and duplicate destinations
Codex review of the move proposal found three gaps in the folder-move
design:
- A skill directory that contains nested skills was planned as one
skill, so descendants would move on disk but keep their old records,
pins and audit acceptances. Plan now expands descendants too.
- Refusals were derived from discovered skills only, so a tracked
checkout with no discoverable skill would move along with its folder.
Preflight now walks the whole tree for checkouts and links.
- Two inputs converging on one destination path were treated as a
forceable flat-name collision and left the batch half-moved. They are
now a separate, non-forceable duplicate_dest refusal.
* fix(proposals): keep nested records through reconcile, warn both filter directions, refuse overlapping roots
Second Codex round on the move proposal found three more gaps:
- The reconcile walk stops at an installed parent and then prunes every
record it did not see, so a moved nested skill would lose its record
and audit acceptances. Reconcile must keep records below a live
installed record.
- Target filter warnings only covered rules that stop matching. A move
into a folder that an exclude pattern covers (or an include newly
adds) changes sync output too, so compare the effective filter result
before and after.
- Batches such as `move foo foo/sub archive` planned the same subtree
twice. Overlapping source roots are now refused before any rename.
Without written rules, new work kept adding entry-point tests that
repeat package rules and rendered UI tests for cosmetic behavior. State
where a rule is tested, what handler/CLI/integration tests cover, and
which dashboard tests to write.
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
Review of the guidance move found no defect, but three things made the new
seam harder to read than it has to be:
- The plan and apply handlers nested the success path inside `if err == nil`
with one shared error write. They return early again, like the handlers
next to them.
- The handler derived the scope from IsProjectMode() while GuidanceInput
derived the same value from ProjectRoot. Scope() is exported so the status
response and the block inspection read one value.
- (*Server).memoryInstructions had become a one-line passthrough with a
single caller, so it is inlined.
GuidanceReader is now exported, so its doc states what the server always
guaranteed: at least one source, and Dest indexing one of them. The
ApplyGuidance and Config comments say what callers need to know, and the
architecture paragraph is rewrapped.
No behaviour change.
Refs #450
The uninstall sequence (move-out check, dirty check, move to trash,
.gitignore cleanup, metadata prune) lived in the CLI and in three
dashboard handlers, and the copies had drifted: three different prune
rule sets, no dirty check on DELETE /api/repos, and no metadata prune
at all on DELETE /api/resources, which left the removed skill's entry
and target overrides in .metadata.json.
internal/uninstall now owns the sequence for resolved items. Callers
keep name resolution, prompts, output, HTTP status and oplog.
MetadataStore.RemoveByNames is the single prune rule set: an entry goes
when it is a removed path or lies below one, by key or by Group, plus
the tracked legacy group of a top-level repo.
Behaviour changes, all intended:
- DELETE /api/repos/{name} refuses a dirty tracked repo, or one whose
git status cannot be read, with 409 unless ?force=true.
- DELETE /api/repos/{name} refuses a followed link whose folder is
itself a skill, like every other uninstall path.
- DELETE /api/resources/{name} prunes the skill's metadata.
- Removing top-level "foo" no longer prunes "frontend/foo" (CLI), and
removing repo "_team" no longer prunes the untracked group "team"
(dashboard batch); both dropped metadata of skills still installed.
Refs #442
The i18n module guaranteed parity between locales and left the rest to
callers. `t` takes any string, so a mistyped key rendered as the key and
nothing caught it; internal/plugin sends translation keys that nothing
compared with en.json; and about 130 call sites chose between `.one` and
`.other` by hand.
- keys.test.ts reads the dashboard source and fails on a key that is not
in en.json. A key built at run time is checked by its fixed head, and a
string that only looks like a key is named in a short documented list.
- internal/plugin/i18n_keys_test.go fails when a "plugins." literal the
package emits is missing from en.json; the one key completed at run
time lists its values.
- plural(key, count) in the i18n module picks the form, with the same
"exactly 1" split the call sites used, so every rendered string is
unchanged. No locale file changed.
Refs #448
Updating a tracked repo is a fail-closed sequence: check for uncommitted
changes, pull, audit what was pulled, reset to the pre-pull commit when the
audit blocks. The CLI had two copies of the pull flow, the dashboard a third,
and the audit gate existed four times (CLI, dashboard, and two in install).
The copies had drifted: the dashboard ignored a failed git status and never
refreshed root-skill metadata, and each copy worded a failed rollback on its
own.
update.TrackedRepo now owns the sequence and returns a structured result;
install.AuditGate is the only gate. internal/update is a new package because
internal/git already imports internal/install. The CLI keeps its spinner,
rows and the "Apply anyway?" prompt, passed in as one Confirm callback; the
dashboard turns the result into its JSON item.
Behaviour changes, all towards the stricter copy:
- the dashboard reports a failed git status as an error instead of pulling;
- the dashboard refreshes root-skill metadata after an update, like the CLI;
- a prompt that fails rolls the update back instead of leaving it applied;
- the rollback-failed and rollback-unavailable texts come from one place.
Accepted findings stay as they were (CLI honours and records them, the
dashboard does neither), now as the explicit AcceptedFindings input. The
"blocked and rolled back" texts are unchanged per caller because the
dashboard matches on them.
Refs #443
The guidance rules (status ranking, mixed_modes, multiple_blocks, the plan
token, and check -> backup -> recheck -> write) were methods on *Server, so
they could only be exercised over HTTP with a configured server, and the
recheck after backup depended on callers using the right helper.
internal/memory now owns status, plan, and apply behind GuidanceStatus,
PlanGuidance, and ApplyGuidance. The server resolves targets, assignments,
shared extras, and project reach into a GuidanceInput, syncs shared copies
through a callback after each written file, and maps results to JSON. Only
ApplyGuidance reaches the write steps, so the recheck cannot be skipped.
Note creation and guidance creation shared a copied O_EXCL create with
delete-on-failure; both now use createExclusive. Each keeps its own
behaviour: notes still chmod to 0644, guidance still creates parents.
HTTP routes, shapes, status codes, and error codes are unchanged. The raw
write allowances follow the calls to their new functions.
Refs #450
Which queries a mutation makes stale was knowledge each caller carried:
139 invalidateQueries calls in 48 files listed keys by hand, the lists
differed, and some used hand-written prefix keys such as ['sync-matrix']
that queryKeys could not express.
The key list of every event now lives once in lib/queryEvents.ts and call
sites say what happened: invalidate(queryClient, 'skillsUninstalled').
The three existing helpers (invalidateAfterSync, refreshTargets,
refreshInstructions) became events too. queryKeys gained the prefix keys
(syncMatrixAll, diffAll, logAll, logStatsAll) and every query key that was
written inline, so no key literal remains outside queryKeys.ts.
No call site's invalidated set changed. Each event is tested once against
a real QueryClient, including which queries it reaches by prefix.
Refs #445
Pass value-taking options to the shared mode parser so command values cannot select a scope or create false mode conflicts. Preserve the end-of-options delimiter and reuse it for MCP command arguments.
Cover existing CLI entry points, including audit rules, enable/disable, and backup deletion, with regression tests and document the parser contract.
Source links disappeared from the dashboard when discovery found no skills, leaving empty or unavailable folders without an Unlink action. Return first-level links independently with their follow-policy availability and preserve their groups in every skills view, including empty unlink confirmations.
Cover empty and dangling inventory, disabled following, and empty-link removal in list, cards, and tree views; document the API field and rendering.
* feat(omp): add native Oh My Pi support
Support OMP skills, instructions, MCP, code hooks and guarded extension selection using its verified native contracts rather than Pi aliases.
Add reviewed marketplace plugin installation, cache-preserving scoped removal and fresh-cache reinstall. Align Pi/OMP extension cards and localize inventory guidance while keeping native ownership and read-only constraints.
Include regression coverage, isolated E2E runbooks, documentation and verification policies. Native plugin updates and unverified runtime platforms remain guarded.
Refs #409
* fix(omp): guard runtime ownership and localize inventory guidance
Native install recursively replaces its runtime destination, and a missing cache manifest could leave the real runtime link installed after removal. Bind and recheck an absent reviewed destination, and recover removal names only from verified scope-local links.
Localize fixed backend reasons and warnings across dashboard locales while preserving paths and external diagnostics. Restore the reviewed source in the existing reinstall fixture and record real native verification.
* fix(omp): allow forgetting natively removed plugins
A different plugin can retain node_modules after native uninstall deletes this installation. When its cache, registration and known runtime state are absent, forget only the stale Skillshare binding instead of demanding runtime-name proof for unrelated files.
Extend the existing removal recovery test with a failing-before-fix case that verifies the binding is dropped and unrelated native lock bytes and plugin files are preserved. make check passes.
A root skill discovered through a followed source link occupies the link entry itself. Moving that entry to trash silently unlinks every nested skill instead of removing one skill.
Share the root-skill guard across single and batch API handlers and CLI uninstall, report the unlink action clearly, and preserve explicit unlink and nested skill removal.
Replace the removed linked badge description with the link icon, muted target path, and hover or focus unlink icon used by the dashboard. Keep the tree detail-pane behavior explicit.
Pass the current global or project follow policy to skill trash restore so the dashboard matches CLI behavior and can restore into followed link targets. Keep disabled-link refusals and agent restores unchanged, and cover both scopes and link naming styles with handler regression tests.
Resolve followed skills with the discovery policy before applying tracked-repo restrictions or prefix-based batch routing. This lets dashboard uninstall move only the selected skill from a followed checkout while retaining managed-repo protection and logical trash restore paths.
Add single and batch handler regressions for global and project modes, skipped links, and tracked-repo members; document the API behavior.
Keep followed checkout updates in the Updates tab and expose unlink as a single quiet icon in group headers. Remove unused checkout metadata and linked-group menus while preserving the confirmation and single-skill uninstall flows.
Match tracked repository affordances without offering whole-repo uninstall for followed user checkouts. Resolve checkout identity for Update repo, keep linked skill uninstall scoped to the selected skill, and explain target preservation, sync impact, and trash recovery before destructive requests.
Expose policy-resolved first-level link identity on listed skills so linked groups can be distinguished from ordinary folders without another request. Add a confirmation that preserves the target folder and explains trash recovery, then refresh the dashboard after unlinking.
Let dashboard users add local skill folders through the shared source-link guards without creating links by hand. Keep follow_source_links opt-in in the current mode and expose link removal through the API with recoverable trash semantics.
Cover handler behavior, request shapes, dialog errors, and all dashboard locales, and document the endpoints in the API audit topic.
follow_source_links only follows a link the user made by hand, so the
guards (source or parent, sync target overlap, link chain) surfaced in
doctor only after the link existed, and Windows users had to pick
between a junction and mklink /D themselves.
link checks the prospective target with Follow.Allow, which shares the
guard rules and reason strings with the walk, before creating a junction
on Windows or an absolute symlink elsewhere. It warns when the target is
not a git checkout and leaves follow_source_links alone unless --enable
is given. unlink moves only the link entry to trash, the same way
uninstall removes a first-level link, so the target is never touched and
trash restore undoes it. internal/sourcelink holds both operations so the
dashboard can call them too.
Refs #419
Separate user-owned followed Git checkouts into linked_repos in check and resources responses so the Updates tab renders information instead of tracked errors or retry actions. Refuse dashboard updates before Git runs, including force retries and descendant skills, while leaving managed repositories unchanged.
Keep one follow policy per list and status operation and report its skipped links without hiding healthy inventory. Send JSON diagnostics to stderr and expose sourceLinkWarnings in the dashboard resources response, reusing the sync warning formatter.
Record how the junction case is verified through the UTM guest and point the testing topic and the config reference at it, replacing the 'not yet verified' caveat now that the ARM64 run passed.
Interrupted Git operations left 32 GB of tmp_pack files in .git, which
filled the disk. Document how to find and clear that and the other
ignored verification leftovers without pruning shared caches.
Wave 2 made writes behind a followed first-level link land in the real
checkout, so the "current limits" paragraph written for the discovery-only
version was no longer true. Replace it with the actual write semantics, add
the walker decision flowchart the design discussion settled on, and record
the sourcefs/sourcewalk policy contract in the CLI development topic.
Refs #274
Use ssenv and explicit worktree and binary paths for parallel CLI verification. Reserve temporary containers for mount or runtime isolation needs and avoid duplicating unnecessary frontend dependencies and Go build caches.
The polling instructions read as applying only after an `@codex review`
comment, so an agent could open a pull request and hand Codex's initial
findings back to the user instead of following them up itself.
Implements rollout steps 1 and 2 of proposals/274-skillfollow.md. Nothing follows a link or parses .skillfollow yet.
Step 1: doctor reports first-level symlinks and junctions in the skills source as info, in text and JSON, without following them.
Step 2a: internal/sourcewalk owns traversal of the skills source. Thirteen raw walks move onto it with identical behavior, and a go/parser ratchet fails the tests on any new unclassified walk.
Step 2b: internal/sourcefs wraps an os.Root opened at the skills source and refuses any write whose path has a link component. The six write sites the proposal found in review move onto it, including the cross-device copy fallback, and a second ratchet covers raw os write calls. The only behavior change is that an explicit path crossing a link inside the source is refused instead of written through.
Refs #274
Release-As: 0.24.3
- The LEARNED.md starter said to update only on request, which
contradicted the active memory guidance; it now defers to the guidance.
- website/AGENTS.md hardcoded a command-page count that had drifted and
omitted the Desktop App navbar item.
- wiki/testing.md gave `bash -c` and `bash -lc` as the rule in different
places; it now says which container needs which, and why.
- Migration-relative wording in wiki/architecture.md and an unexplained
"Always use" in the targets reference are stated as current rules with
their reason.
Config files written by the CLI were marshaled straight from the Go
structs, so sections ran together and plugins, hooks and mcp landed
above sources. Programmatic writes of the global and project config
now sort sections like the dashboard and separate them with a blank
line, keeping the original order when sorting would break a YAML alias.
Also documents one worktree and throwaway container per branch task.
Release-As: 0.24.1