Adding a plugin from a Git URL to Oh My Pi on macOS was always blocked
with "linked or non-directory OMP path component", which the dashboard
shows as "The reviewed OMP runtime destination must be verified empty".
The source is cloned into os.MkdirTemp, which lives under /var/folders,
and /var links to /private/var. The OMP install check reads the source's
package.json through ompSafePath, which rejects any linked component, so
Skillshare's own temp clone failed it. A local source under a linked
parent failed the same way.
acquireRef now hands callers the resolved root of the temp clone and of
a local directory, while the recorded source stays as given. Links inside
the source are still handled as before; cleanup still removes the
original temp directory.
* fix(plugin): accept OMP 18.6.1 and later instead of exactly 18.6.1
The OMP adapter pinned the native version to exactly 18.6.1, so every
newer release was refused: the dashboard reported "Installation requires
OMP 18.6.1" for a user on 18.8.7, and removal and standalone extension
selection were blocked the same way.
The sources the adapter depends on (marketplace cache/manager/types,
plugin-cli output, the project anchor in discovery/helpers.ts, extension
loading and the pi-natives file lock) are unchanged from v18.6.1 through
v18.10.2, so the pin is now a minimum. The install, removal and selection
gates share one check built on the existing plainVersion helper;
unparsable, prerelease and older versions stay blocked.
Messages, all dashboard locales, the plugin reference docs and the
built-in skill now state 18.6.1+. A test keeps the English guidance text
in en.json equal to the Go strings the dashboard matches on.
* fix(docs): record the OMP 18.6.1+ boundary in the cache safety report
The safety report still said installation is allowed only for OMP 18.6.1
and that an unknown version blocks it, which no longer matches the gate.
It now states the minimum, which versions stay blocked, and the source
comparison and native 18.8.7 run that justify replacing the pin. The two
OMP runbooks no longer require exactly 18.6.1.
* fix(mcp): turn off a global server in OpenCode 2 projects
OpenCode 2 replaces a global server with the project entry of the same
name, and silently drops an entry without its type and command or url.
The project switch Skillshare wrote, {"enabled": false}, therefore left
the server on in OpenCode 2, while OpenCode 1 merged it by field.
Checked against OpenCode 2.0.27 and 1.19.0: OpenCode 2 still reads the
flat mcp.<name> shape and turns enabled: false into disabled: true, and
both read an entry with type and command or url. So no version detection
is needed. When the global config under mcp.projects defines the server,
the OpenCode switch now carries its type and executable, or its url
without the query, next to enabled: false. Args, env, headers and the
query stay out of a project file that may be committed, as the Pi switch
once did. A switch that already carries the endpoint keeps it when the
global server later leaves the config, unless it was edited since sync.
In project mode, or when the global config lacks the server, only the
bare switch can be written; the plan now says OpenCode 2 ignores it, in
the CLI and the dashboard.
Import also reads OpenCode 2's mcp.servers section, which its own
`mcp add` writes, instead of reporting it as one broken server.
Refs #538
* fix(mcp): keep secrets in a URL path out of the OpenCode switch
The OpenCode switch dropped only the query of the global server's URL.
Some MCP services put the key in the path instead, and an absolute
executable path names the user's home directory; both then reached a
project file that may be committed. A disabled server never starts, and
OpenCode 2 accepts any type with a command or url, so the switch now
carries only the executable's name or the URL's scheme and host.
Refs #538
* fix(mcp): skip OpenCode 2's timeout defaults when importing
OpenCode 2 keeps server defaults in mcp.timeout next to mcp.servers, so
import listed timeout as a broken server. It is dropped only together
with a recognized servers section, where it cannot be a v1 server.
Refs #538
* fix(mcp): read an empty OpenCode 2 servers section
An empty mcp.servers object was not recognized as OpenCode 2's section,
so import listed servers and timeout as broken servers instead of
finding no entries. The dashboard help for disabled also still said only
the switch is written, which is no longer true for opencode.
Refs #538
* fix(mcp): keep an OpenCode endpoint switch next to its own fields
The kept switch was matched by the whole native entry, so a field only
OpenCode reads, such as a per-server timeout, made the next sync drop the
type and command or url once the global server left the config, and
OpenCode 2 then ignored the switch. Match the fields Skillshare manages.
Refs #538
* fix(mcp): show the OpenCode endpoint switch in the dashboard preview
The dashboard renders a project's server alone, so the preview never saw
the same-named global server and showed the bare switch that OpenCode 2
ignores, while sync wrote the switch with the endpoint. Resolve the
global server the way sync does.
Refs #538
* fix(plugin): read OpenCode 2's version output
OpenCode 2 prints "opencode v2.0.27" for --version, while OpenCode 1
prints only "1.19.0". The adapter trimmed a leading "v" only, so plugin
sync to OpenCode 2 failed as an unrecognized version and imported
updates were skipped as if OpenCode 1 had no update command. Read the
last field, and decide the update check with the same parser.
Refs #538
An imported plugin's native marketplace decides the version an update installs, so the update check now applies only to bindings with a reviewed source and an import records the version installed. Importing an imported plugin again also records the installed version; a source binding keeps the version Skillshare reviewed.
Refs #535
Updating several plugins from the dashboard was slow: each plugin took a
preview and an apply round trip, apply previewed again, every plugin cloned
its source on its own, and config reloaded after each one.
The dashboard now sends the picks to POST /api/plugins/update/stream. The
server holds the plugin lock once, previews each plugin once and applies that
plan, lets plugins from one source share a single read of it, and reloads
config once at the end. POST /api/plugins/update/stop lets the plugin under
way finish and starts no other; a closed page does not cut an install short.
The check-to-update safety moves to the server: an Agent whose source moved
since the check, or that no check showed, installs nothing. npm packages in
Pi are exempt, since Pi installs whatever npm names latest; the old
client-side comparison flagged them as moved on every update.
The update dialog follows the Skills update page, shows per-Agent old -> new
versions, fails rows a broken stream never finished, and lists every plugin
with its Agent and version when all are current.
Release-As: 0.25.6
* feat(ui): check plugin updates on load and update them in one batch
Updating plugins took four steps each: press Check updates, press Update
on one row, review a second preview, apply, then check again because the
list did not know the update had happened. Six plugins meant six rounds.
The page now checks once in the background when it opens, so rows show
their new version without a click. The header's Check updates button
becomes "Update N" when something is behind and opens one dialog: every
plugin with an update is ticked, a plugin on several Agents is one row
whose Agents can be left out, and imported installs Skillshare cannot
update sit apart. One button runs them in turn through the existing
preview and apply, so no new API is needed. Progress, results, errors
and per-Agent retry stay in the same dialog, and the applied results
update the page state, so nothing has to be checked again.
The count and the dialog both come from what is still behind after the
last check, limited to Agents that can be updated from here, so a
partial batch is not offered again and Agents with check but no update
(OMP, Antigravity CLI, Grok) do not inflate the count.
* fix(ui): keep batch plugin updates to what the user reviewed
Codex review of #527 found four ways the batch update could act on
something other than what the dialog showed:
- A source that moved between the background check and the click was
installed straight from the fresh preview, so a newer version or new
components went in unreviewed. The batch now compares each fresh plan
with the version and commit the check showed, applies nothing when
they differ, says so on the row, and runs a quiet check so the next
dialog shows the new version for review.
- A retry of the failed Agents that failed again lost that subset, so a
third try re-sent Agents that had already updated. A failed row now
always keeps the Agents it failed on.
- An update the server skipped (its marketplace gone, say) counted as
done: the row said updated and the plugin left the update count.
Skipped outcomes now keep the Agent pending and show as skipped.
- A plugin removed after the check stayed in the update count and
failed with "not found" when picked. Pending updates now drop any
plugin or Agent binding the inventory no longer has.
* fix(ui): show a manual check's result and each Agent's own version move
Codex's second review of #527 found two gaps:
- A check the user pressed that found nothing to update from here (only
imported installs, or everything current) opened nothing, so the
"update inside the Agent" guidance and the up-to-date result never
showed. The check now reports whether it came back, and a check the
user asked for opens the dialog; the background check on load stays
quiet.
- A plugin on several Agents showed one version move built from the
first Agent's installed version and the first new version found, which
can mix Agents that come from different sources. The row now shows one
move only when every Agent makes the same one, and each Agent's own
move otherwise.
* fix(plugin): carry the source digest on a check so a batch update can tell it moved
Codex's third review of #527: a local-directory plugin has no commit,
so when its files changed again after the check while the manifest
version stayed the same, the batch's version and commit comparison
passed and the fresh preview was applied without the user seeing it.
The check copied the installed binding, so its digest was the one
already installed, not the source's. It now sets the discovered digest
alongside the version and commit, as the update preview already does.
The dashboard keeps that digest with each update it found and refuses
to apply when the fresh preview's digest differs, marking the row as
changed since the check. An npm package has no digest and is unaffected.
* fix(ui): list blocked check results and drop unticked Agents from the batch
Codex's fourth review of #527:
- A check can mark an Agent blocked when the updated source no longer
offers a compatible plugin for it. The update dialog listed only
updatable, native-check and up-to-date results, so it hid the block
and could say everything was current. Blocked results now get their
own section with the reason, and the empty state says nothing can be
updated from here rather than that everything is up to date.
- An Agent unticked after the check stayed in the pending updates;
picking it produced a no-op that the batch reported as updated.
Pending updates now skip bindings with sync off, as the row's own
behind count already does.
* fix(ui): count partly skipped plugins as updated and surface a failed re-check
Codex's fifth review of #527:
- When one Agent of a plugin updated and another came back skipped, the
row was marked skipped, the summary counted no update, and the Agent
that did update was not shown. A row now names which Agents updated
and which were skipped, and the summary counts any plugin with an
updated Agent.
- After a source moved since the check, the follow-up check ran quietly,
so if it failed the stale update stayed offered and every retry hit
the same skip. The moved plugin now leaves the found updates, and the
follow-up check reports its failure like any check the user started.
Saves follow a symlinked config.yaml to its target so dotfile links
survive. For a project config that also lets a cloned repository point
.skillshare/config.yaml at any file the user can write, including one
that does not exist yet, and every project save, plugin change, hooks
change, MCP change, or legacy migration would then rewrite or create it.
All five writers now resolve the destination through one helper,
utils.ConfigWritePath. A global config keeps writing through its link.
A project config must resolve inside its project; otherwise the save
fails and names the link so the user can replace it with a regular
file. The one-time legacy migration skips persisting instead, and the
next save reports the link.
FoldHomePathWith writes ~\plug on Windows and a bare ~ for the home
directory itself, but canonicalSource only expanded ~/, so with
preserve_tilde_on_save those sources came back relative on the next
load and update or sync could not resolve them.
When config.yaml is a symlink whose target does not exist yet, as a
dotfiles manager may create before first use, EvalSymlinks fails and
the save renamed over the link again. Follow the link to its intended
target so the first save creates that file, as Config.Save already does.
Refs #460
Plugin saves wrote a temp file next to config.yaml and renamed it over
the path. When config.yaml is a symlink into a dotfiles folder, the
rename replaced the link with a regular file, so the shared config
never received the change and the machine silently forked. The plugin
writer also skipped preserve_tilde_on_save, saving sources as absolute
home paths. The one-time legacy target migrations for global and
project configs used the same temp+rename and replaced links as well.
Plugin sources were compared as raw strings. The snapshot owner marker
records the absolute path from add, so a config that spells the same
folder as ~/... failed every update with "snapshot ownership conflict".
For Agents whose snapshot location hashes the source, the new spelling
silently created a second snapshot instead.
Resolve the symlink before writing, normalize sources once where the
plugin section is decoded so owner checks and snapshot paths agree,
and fold sources back to ~ on save when preserve_tilde_on_save is set.
Refs #460
Refs #461
Review showed three ways a missing key still passed:
- t('sync.changes') passed because the counted base was accepted for
any string; a string passed straight to t() must now be a key itself.
- plural('typo.key', n) was only seen when its first word was a known
namespace; strings passed straight to plural() are now checked for
both forms.
- The Go test listed the values that complete "plugins.problem." by
hand, so a new case in automationProblem passed; the values now come
from that function's switch.
The files of config paths were also skipped by every check; they are
now skipped only where config paths look like keys, so their t() calls
are checked again.
Refs #448
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
Update checks compare the source digest, not the version, so a repo that
pushes commits without bumping its manifest (ponytail ships dozens a day)
showed "4.12.0" with an Update button and looked like a stuck update.
Check now returns the source commit, and the version tag adds
"old → new" commits when the version itself did not move.
* 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.
Codex review on #400:
- The private-registry check read npm's config by hand, and each round
found another place npm reads it from: the lowercase environment, the
global npmrc, and now a globalconfig or prefix set in the user .npmrc.
Rather than chase npm's rules, the check now asks npm itself:
`npm config get registry @scope:registry` in the folder Pi runs npm in.
Any registry other than npmjs, or npm not answering, leaves the package
to Pi. The hand-written .npmrc reading is gone.
- When an apply stopped before any Agent ran, such as on a stale preview,
the result had no outcomes, and the page took every planned update off
its list, hiding the row's Update action. Only an Agent the result
reports done comes off now.
Codex review on #400:
- The Plugins page kept only one version per plugin from a check, and
only when the change had one, then hid the row's Update button when an
Agent already had that version. A source plugin whose content changed
without a version bump, or whose manifest has no version, so lost its
Update action once the check dialog closed. The page now keeps what the
check found per plugin and Agent, with or without a version, and an
applied update takes those Agents off; the version only shows what the
update brings.
- The private-registry check did not read npm's global config, so a
registry set only in $PREFIX/etc/npmrc still sent the package name to
npmjs. It now reads the file npm_config_globalconfig names, else
$PREFIX/etc/npmrc with the prefix from the environment or the npm on
PATH, and leaves the package to Pi when it can't tell where that file
is.
- Reading the package version made a package.json whose version is not a
string fail to decode as a whole, which dropped its pi manifest and the
extensions it declares. The version is decoded on its own now, so an
odd one is only skipped.
Codex review on #400:
- A package added with a range such as npm:demo@^1.2.0 was compared with
npm's unrestricted latest. With latest at 2.0.0 and the newest matching
release at 1.9.0, check kept offering 2.0.0, which Pi never installs
because it keeps the configured range. Only a package added without a
version follows latest now; one with a range or a tag is left to Pi, as
a pinned one already was.
- The private-registry check read the user .npmrc path only from
NPM_CONFIG_USERCONFIG. npm reads its environment case-insensitively, so
a lowercase npm_config_userconfig pointing at an .npmrc with a private
registry was ignored, and the package name went to npmjs. Both
spellings are read now.
The plugin reference, its translations and the built-in skill say which
packages the check leaves to Pi.
Codex review on #400:
- The update check asked the public npm registry about every unpinned Pi
package, even when the environment or an .npmrc sends the package, or
its scope, to another registry. A public package with the same name
could then hide a real update or offer an unrelated version. The check
now reads the npm registry settings for the package and its scope from
the environment, Pi's npm folder and the user .npmrc, and falls back to
checking in Pi when any of them names another registry.
- The row's Update button disappeared as soon as any Agent had the
version the check found, so with one Pi target behind and another up to
date, the one behind lost it. The row now compares each selected Agent
with that version on its own, and updates only the ones behind.
The Plugins page listed every binding a run touched, one row each, so a
single failure sat among a dozen "unchanged" rows, and the preview did
the same. Failures now come first, Agents that ended the same way share
a row, and what stayed the same folds into one Unchanged line that opens
to one row per plugin, with Pi packages apart. Logos stay only on the
rows that need a look.
An npm package in Pi recorded no version and no source, so it showed no
version and Check updates could only say to check in Pi. The installed
package.json now gives its version, on the Plugins page and in a Pi
target's Extensions tab, and check compares that version with npm's
latest. A newer version gets an Update button on the row and in the
check's result; it updates that one package with pi update <source>,
not pi update --extensions, which would also update packages the
preview never listed. update leaves a package already at npm's latest
alone, and records no version, so the dashboard keeps reading what Pi
installed. The two check messages that were English in every locale now
have keys.
Release-As: 0.24.5
Codex review on #399: switching an extension of a global package in a
project writes a project override, which the project view reads in its
project-entry loop. In project mode that loop looked only at the
project's config, so the override lost the global plugin's name.
A project entry now falls back to the global config's owners when the
project config has none.
Codex review on #399: a dashboard started in project mode gives the
plugin service the project's config, so a global Pi package that
Skillshare installed, inherited by the project, found no owner and its
card fell back to the state path again. The earlier test kept
ConfigPath on the global config and missed this.
The service now takes the global config path in project mode, and the
project view looks up inherited global entries there.
Codex review on #399: an entry with a readable source but a value
Skillshare can't edit, such as "extensions": null, returned early in
both the global and project views before ManagedBy was set. Its card
then fell back to the plugin's local state path again.
The lookup now runs for every entry whose source can be read, before
the early return.
Codex review on #399: piProjectState never set ManagedBy, unlike
piGlobalState. A project's Extensions tab (<project>@pi, or pi in
project mode) therefore titled a plugin Skillshare installed by its
state path and dropped the "Managed by Skillshare" tag, for both the
project's own entries and the global entries it inherits.
The lookup by source, then by resolved local path, moves into
piManagedBy and runs for every package of both views.
The macOS finder tests stub PATH, but the candidate list still checked
/opt/homebrew/bin/codex, /usr/local/bin/codex and /Applications on the
real filesystem. In the devcontainer /usr/local/bin/codex exists, so it
won over the app bundle each test set up and make check failed there.
The finder now joins those absolute paths onto a root field, "" on a
real machine, and the tests point it at an empty temp dir.
Adopting the portable root plugin.json replaced the whole Codex target
info, dropping the interface.logo read from .codex-plugin/plugin.json.
Codex still overlays that native manifest, and the portable one rarely
names a logo, so the add/list UI fell back to a generic icon.
Refs Codex review on #396.
A portable root plugin.json without a version left the Codex binding's
version empty, which skips the post-update version check, so an update
could advance the snapshot while Codex still ran an older copy. Codex
reports 1.0.0 for such a manifest, so expect that instead.
Refs Codex review on #396.
Codex 0.159 installs from a portable root plugin.json (agent-plugins.org
schema) ahead of .codex-plugin/plugin.json, and reports 1.0.0 when that
manifest has no version. Skillshare preferred .codex-plugin/plugin.json,
so updating a package that ships both (ponytail 4.10.3) failed with
"native update returned version 1.0.0; expected 4.10.3" on every sync.
* fix(plugin): find the Codex CLI the desktop app ships
Plugin commands ran the bare name `codex` from PATH, so a machine with only
the Codex desktop app, or a scheduled job started with a minimal PATH,
reported the CLI as missing. The account-only `cli` override did not help:
the default codex target cannot set it, and an absolute path in a config
shared between macOS and Windows is wrong on one of them.
When `codex` is not on PATH, look in Homebrew's folders and the CLI the app
installs (Codex.app on macOS, %LOCALAPPDATA%\OpenAI\Codex\bin on Windows).
SKILLSHARE_CODEX_CLI overrides the search on one machine without touching
the synced config. The missing-CLI error now lists every place searched, in
the CLI and the dashboard.
Refs #338
* docs(plugin): note where Skillshare finds the Codex CLI
The plugin reference and the sharing guide still said every native CLI must
be on PATH, and the built-in skill told agents a missing CLI has no
fallback. Codex now has one, plus a machine-local override that agents
should prefer over a path in a shared config.
Refs #338
* fix(plugin): look for the Codex CLI inside ChatGPT.app
Since the Codex desktop app merged with ChatGPT, macOS installs it as
ChatGPT.app, so an app-only Mac with a minimal PATH still reported Codex as
missing. Try ChatGPT.app before Codex.app in /Applications and
~/Applications, the order Codex's own desktop-app finder uses.
Refs #338
* fix(plugin): run the packaged CLI entrypoint of the ChatGPT app
ChatGPT Desktop 26.924 removed Contents/Resources/codex and ships the CLI
at the entrypoint its package manifest declares, codex-cli/bin/codex, so a
current app-only Mac still reported Codex as missing. Try that entrypoint
first in each app bundle and keep the old path for earlier builds.
Refs #338
* fix(plugin): find the Codex CLI from the Windows installer
The official PowerShell installer puts codex.exe in
%LOCALAPPDATA%\Programs\OpenAI\Codex\bin and adds it to the user PATH. A job
started without that PATH missed it, the same gap Homebrew's folders cover
on macOS. Try it before the desktop app's copy.
Refs #338
* fix(plugin): keep an account's cli even when it is codex
The Codex search ran whenever the executable was the bare name codex, so an
account that set cli: codex could get SKILLSHARE_CODEX_CLI or the app's CLI
instead, against the documented rule that an account's own cli is kept.
Search only when no account cli is set.
Refs #338
Turning on a file that only a "-path" rule kept off wrote "+path", so the
row ended with a rule pinning it on and a separate Remove rule button that
did nearly the same thing. When removing the row's own rule gives the state
the switch wants, the view now says so (unruled) and the switch removes the
rule. Remove rule stays only where the rule is redundant. The prediction
uses the same rule edit, evaluation and guard as the write, so the preview
and the result agree.
* fix(plugin): trust symlinked ancestors of the Agent config root (#365)
noSymlink walked every ancestor up to / and refused the write if any was a
link, so macOS /var, a home on a linked volume, a dotfiles-linked ~/.pi, or a
PI_CODING_AGENT_DIR / CLAUDE_CONFIG_DIR / CODEX_HOME behind a link failed with
"refusing to modify symlinked native path".
noSymlink now takes the root the user chose (the Agent folder, the project,
or Skillshare's state dir) and checks only the path and the folders between
it and that root, refusing paths outside it. The root is compared lexically
because every checked path is built by joining onto it, so no EvalSymlinks
call is needed. A linked config file or a linked folder inside the root is
still refused, and Windows junctions are still detected via IsLinkMode.
* fix(plugin): bound the noSymlink ancestor walk by its relative depth
The walk stopped when p equalled root, but on Windows Dir spells a UNC
share root with a trailing separator (\\server\share\) while Clean keeps
the root without it, so the loop never ended for a PI_CODING_AGENT_DIR,
project root or StateDir on a share root. Walk up exactly as many levels
as the relative path has instead.
* feat(pi): edit extensions on any Pi at or above the verified minimum
Pi 1.0.1 was released and every user who updated found the Extensions
tab read-only, because the gate matched an exact list of versions. The
package contract has not changed between patch releases, so the gate now
accepts any plain X.Y.Z at or above PiMinVersion (0.99.2), the oldest
version with executed evidence. Older, prerelease or unparsable versions
stay read-only.
The version matrix now also records Pi 1.0.1, which passes every check.
The evidence test requires PiMinVersion to be recorded and every recorded
version to pass, instead of matching an exact list.
The read-only reason is renamed to unsupportedVersion and the API returns
minVersion, so the dashboard says which version is needed.
The Pi MCP project switch keeps writing the global command or url: from
1.0.1 Pi accepts a switch-only entry, but it is skipped by older Pi and
warns at every start where the global config lacks the server. Its
rationale in comments and docs is updated to say so.
* fix(docs): update Pi extensions design records for the minimum gate
Codex review on #367: the design records still described the exact
version list (PiVerifiedVersions, unverifiedVersion, "exact versions
only"), so a contributor relying on them would implement or validate the
old contract. implementation-report.md now states the current gate and
evidence, and phase-0.md gains a dated minimum-version update that the
older paragraphs point to, keeping the history intact.
* chore(design): export dashboard tokens for design tools
Mockup tools read a design system's tokens.json rather than our CSS, so
designs drifted from the dashboard's Clean theme. The script derives the
tokens from components.css, and CI runs its test so a new color
variable without a usage note fails instead of silently going missing.
* feat(plugin): add npm packages from pi.dev through Pi
Pi packages on pi.dev are npm packages, which Plugins could not add:
it only took local or Git sources, so users had to install them with Pi
and import them. `plugin add npm:<package>` now hands them to
`pi install` for Pi targets and records the binding like an import.
Skillshare can't fetch or review an npm package, and Pi runs its install
scripts, so the preview says so, discover refuses npm sources, and only
targets that run pi itself take them. Pi keeps one entry per package
name, so another version replaces it in place and keeps its extension
filters; Skillshare records the new entry so a reinstall restores them,
and refuses a package another Skillshare package already manages, since
removing one would remove both. Pinned versions are skipped on update
because pi update keeps them.
The dashboard accepts npm sources, a pasted `pi install` command or a
pi.dev address, lists only Pi targets, and the Pi Extensions tab links
to adding a package with that target selected. Pi packages show Pi's
mark in the plugin list.
A blocked change now keeps its own reason when the Agent's CLI is also
missing, and the message key always matches the message the CLI prints.
* fix(plugin): keep npm package names and accept pi by path
The default name of an npm package went through the file-path rule, which
cut its last dot segment as an extension: npm:foo.bar was recorded as foo,
and adding npm:foo.bar@1.2.3 later derived foo.bar and was refused as
another package managing the same Pi entry. npm names are now taken from
the package spec.
A Pi account whose CLI is an absolute path to pi was treated as running
another executable, so it could not take npm packages. It now uses the
same path- and platform-aware check as the Pi Extensions tab.
* docs(wiki): add a pull-requests topic for Codex review rounds
Handling Codex's review was only known from practice: confirm each
finding, fix it with a failing test first, reply in its thread, ask for
another review, and follow up until it answers. Writing it down keeps
agents from resolving threads, amending pushed commits, or stopping at
the 👀 that only means the review is still running.
* feat(ui): give the Extras AGENTS.md and Memory tabs one layout
The two tabs shared a page but looked like different products: the header
shifted between them, notes sat in a joined box with an icon toolbar, and
agents showed as an icon strip in a side card. Switching tabs felt jarring.
Memory now follows the AGENTS.md layout: a sticky list column with search,
the note's name with Edit and a More actions menu, the same preview box,
and Use with agents as one row per tool under the note. Long notes open
collapsed like shared files do. The AGENTS.md list gains a search box and
the same sticky column. The Memory tab label stays English in every locale,
like AGENTS.md. Docs and screenshots follow the new layout.
* chore(docs): add a script that serves the Memory screenshot homes
The Memory how-to screenshots came from demo homes that were never
recorded, so reshooting them meant rebuilding the data by hand. The script
rebuilds the fresh and the populated home inside the devcontainer, serves
each behind the Vite dev server, and hands the port back to air afterwards.
* fix(plugin): refuse npm sources Pi can't resolve
validNpmSource accepted specs such as npm:foo/bar, npm:foo@ and names
containing "..", which Pi's own parser rejects. They passed preview, and
Apply recorded a pending binding before pi install failed, leaving an
entry every later sync retried. Validate with the same grammar the Pi
settings reader uses, which also covers a scope without a package name.
* fix(ui): show the default name when adding an npm package to Pi
The name field for an npm source was blank, so it looked like a name was
missing. Leaving it empty records the package name without its version or
scope, as the git flow already shows; the field now shows that name as its
placeholder.
* test(pi): reproduce Windows private registration access failures
Use NTFS DACL assertions instead of POSIX mode bits and verify that broadly accessible existing records cannot be read or silently reused. Keep the Unix mode assertion unchanged.
* fix(pi): reclaim Windows locks and protect private registration state
Anchor stale directories with delete sharing and delete the verified empty Windows directory by handle, keeping inode, lease and racing-owner guards. Create new registration directories with protected owner/SYSTEM DACLs and validate existing owner/access rules without changing their ACLs. Leave Unix permissions and directory-only removal unchanged.
* test(pi): capture Windows lock identity before path replacement
Windows Lstat lazily fetches file IDs by pathname. Snapshot the stale fixture through its anchored handle and add a matching-mtime replacement regression before first verification or release.
* fix(pi): snapshot Windows lock IDs and accept privileged default owners
Capture acquired lock identity through a handle before Windows can lazily resolve a replaced pathname. Elevated tokens can default file ownership to Administrators; permit privileged owners only alongside a verified current-user-only DACL. Refuse unsafe existing registration directories without modifying their ACLs or writing records.
* docs(pi): record Windows acceptance and isolated regression workflow
Document private Windows ACL requirements and refusal without permission rewrites. Add pinned devcontainer kit packaging, token-isolated full/basic acceptance, preserved UTF-8 logs and a reproducible UTM runbook. Preserve the original Unix permission assertion.
* test(backup): isolate explicit agent backup from sync snapshots
Sync automatically backs up target-local agents. Creating the fixture before sync makes the exact-one-backup assertion depend on both commands sharing a timestamp second. Create it after successful sync so the explicit backup is the only operation that snapshots it, without weakening the assertion.
* feat(pi): manage individual extensions per target
Keep package lifecycle in Plugins while letting global, account, and project Pi targets configure individual extensions with reviewed, lossless changes. Project overrides save only project settings and never grant trust.
Bind edits to native version contracts, settings snapshots, source ownership, and discovery; preserve unrelated resources and enforce renewable native locks, stale rejection, and backups. Verify Pi 0.99.2 and 1.0.0 without running extension factories.
Simplify configured-state UI and distinguish Skillshare-managed packages from native tool registrations, with localized guidance and regression coverage. Closes#342.
* fix(pi): address project extensions review findings
Keep exact project delta rules visible when their extension files are missing so users can remove stale rules without changing global settings or trust.
Show the project Extensions tab only when its generated Pi target resolves. Cover missing and agents-only projects, absent targets, and another project target with regression tests.
* fix(pi): recognize official Windows account launchers
Accept pi.cmd and pi.exe account launcher names on Windows, including case-insensitive and native Windows paths, before probing the existing exact-version gate. Preserve POSIX behavior and never probe unrecognized forks or stacked launcher suffixes.
Add platform-name and account-gate regressions covering verified, unsupported, and missing CLI versions. Focused race tests, make check, and Windows amd64/arm64 test cross-compilation pass; real Windows runtime remains unverified because the UTM guest is stopped.
* fix(pi): align rule removal and backup descriptions
Offer Remove rule only for owned exact +/- rules, not project delta globs that the apply path cannot remove. Keep supported glob rows switchable and cover delta and deltaOnly forms with RED/GREEN regressions.
Describe retained artifacts as extension-list and file-hash records rather than complete settings backups in English, all four website translations, and the implementation report.
* fix(pi): discard records for aborted extension applies
Require a validated preview and owned native lock before recording an extension change. Discard only the newly created record if final revalidation, lock ownership, or the atomic settings write fails; retain successful history and propagate cleanup errors with the original failure.
Keep discovery revalidation after record I/O so cleanup does not weaken the write-boundary guard. Cover global and new/existing project settings with nine RED/GREEN regressions, including injected commit failures. Align English and translated docs with the retained-record lifecycle.
* fix(pi): retain anonymous global packages in projects
Keep ordered anonymous registrations separate from the proven-identity map so unsupported query-bearing sources remain visible as redacted read-only packages, rather than being dropped or associated with a project override.
Preserve first-global-entry precedence for known identities. Add RED/GREEN coverage for string and object registrations, query/userinfo redaction, project entries, zero-write refusal, ordering, and continued editing of known packages.
* fix(pi): deduplicate only proven project identities
Exclude unresolved identities from project duplicate tracking. Preserve each anonymous registration as sourceUnknown instead of pretending unrelated query-bearing URLs are the same package.
Cover mixed string/object anonymous project sources, redaction and zero-write refusals, retained anonymous globals, and last-project-entry precedence for known packages. Verify known project entries remain editable.
* fix(pi): guard unresolved package precedence
An unresolved earlier global registration may own a later known package, and an unresolved later project registration may supersede an earlier known entry. Keep those potentially shadowed entries and inherited deltas Unknown/read-only instead of offering edits to ignored registrations. Do not infer identities by dropping credentials or queries.
Cover both ordering directions for global, inherited, project-owned, and delta views with RED/GREEN tests. Execute native query identity and dedupe evidence on Pi 0.99.2 and 1.0.0 through core and CLI-bundle contracts, now 28 scenarios and 87 assertions each. Align docs and evidence while retaining independent anonymous entries.
* fix(pi): propagate unreadable source ambiguity
Treat badSource entries as unresolved instead of letting an unreadable source expose editable later global registrations. Preserve the same uncertainty in inherited globals and before unreadable later project entries.
Add twelve global/inherited/project regressions for null, empty, whitespace and invalid UTF-8 sources. Native core and CLI-bundle evidence confirms invalid UTF-8 can decode to and own the same identity as a later valid source on both supported Pi versions; the executed contracts now pass 29 scenarios and 90 assertions.
* feat(pi): preserve filtered imports and reclaim stale locks
Adopt native object registrations without changing their configured resources. Keep lossless entries in scoped private state, retain current rules across lifecycle operations, restore filters before native installation, and preserve pending retries after failures.
Reclaim only unchanged empty directories after the native stale threshold, retaining renewed, replaced, nonempty and non-directory locks. Verify exact supported Pi contracts and keep unsupported cases read-only.
* docs(pi): record filtered import verification after branch sync
Document the current merged UI test count and repeated checks while retaining the native and visual verification limitations.
* docs(pi): clarify stale acquisition exception
Keep the lock safety summary consistent with the approved acquisition policy while preserving strict ownership for renewal and release.
* fix(pi): keep legacy npm fallback installations unknown
A missing user-scoped managed npm root cannot prove a package is absent because native Pi may use a legacy global npm or pnpm root. Keep it read-only without probing host-global state, while retaining missing-state behavior for project npm packages.
* fix(pi): preserve project precedence and batch restore ownership
Retain empty winning overrides when deleting them could expose older resource filters. Advance filtered restore batches only to exact writes made by the same Apply, while rejecting external changes and retaining failed native retries. Match Pi strict-false autoload semantics instead of treating null as a delta.
Removing a Skillshare marketplace from Claude passed --scope user (or
project). Claude refuses a scoped removal when the marketplace is only
recorded in known_marketplaces.json, which happens once synced settings
drop its declaration, so cleanup failed on every sync. Remove it without
a scope: the name is unique to the binding and its root is checked
first, so every scope's copy is Skillshare's. A name declared at
several paths is reported as a conflict and nothing is removed.
Claude also reads a skill folder that has a plugin manifest as
<name>@skills-dir and loads only one plugin per name, so installing a
Claude plugin named like such a skill made Claude report the skill as
not loaded with no warning beforehand. The install preview now says
which copy Claude will load.
Validation: make check, the UI vitest suite and tsc in the devcontainer;
the cleanup and clash tests fail without the change. Claude 2.1.287
rejected the scoped removal and accepted the unscoped one against a copy
of an affected config.
Release-As: 0.23.4
Managed Claude/Codex marketplaces were named skillshare-<hash>, which
says nothing in native marketplace lists, and they stayed registered
after the plugin was excluded or removed. Cleanup ran only after a
native uninstall, so a plugin already removed in the Agent, a failed
first install, or a failed cleanup left the marketplace behind for good.
Name new marketplaces skillshare-<plugin>-<hash>, and remove the one
Skillshare registered whenever its plugin is excluded or removed, also
when the plugin is already gone. Cleanup checks that the marketplace is
registered from Skillshare's own state directory, re-lists afterwards,
and stays pending for the next sync on failure or while the marketplace
list cannot be read. An update registers a missing managed marketplace
again.
An imported plugin reinstalls from its native marketplace. When that
marketplace is gone (for example when synced settings dropped it), sync
failed with a generic "command failed". Skip that target with the reason
instead, so other Agents still sync. Skillshare does not move an import
to its recorded source on its own: that would install a distribution
and version nobody reviewed for that Agent.
Failed outcomes now carry their translation key, and the dashboard
shows each distinct failure once instead of the joined English text.
Validation: make check and the full UI vitest suite in the devcontainer,
tsc and eslint; the new Go tests fail without the change. Claude and
Codex marketplace removal was checked against the real CLIs in
isolated config directories.
Release-As: 0.23.3
A plugin installed on several Agents could not be updated anywhere when
one of them could not take the update: a plugin disabled in Codex or
Copilot, or an imported package, blocked the whole plan, so Claude
stayed behind too. These targets are now skipped with the reason, and
the rest update. A skipped update stays pending, where the previous
noop cleared it and left Skillshare believing a stale Agent was current.
Two of those were not real limits: an imported Codex plugin now updates
through `codex plugin marketplace upgrade`, and an imported Pi package
in global mode through `pi update`, which keeps its settings entry.
Skip reasons carry a message key so the dashboard translates them, and
a plugin whose Agents hold different versions shows each instead of
none.
Refs #309
Codex has no plugin update command, so Skillshare refused to update a
Codex binding and left that Agent to be maintained by hand. Codex's add
replaces an installed plugin with its marketplace's current copy
atomically, so an update now refreshes the reviewed snapshot, adds the
plugin again, and checks the installed version.
Add always enables the plugin, so one disabled in Codex is skipped with
the reason rather than turned back on. An imported binding is skipped
too, since Codex upgrades Git marketplaces itself. Both are skipped, not
blocked, so the plugin's other Agents still update in the same run.
Refs #309
Pi forks such as omo keep Pi's directory layout and commands but ship
their own executable, so an account target pointed at one still ran
`pi` for plugin commands. A shell alias cannot help, since Skillshare
runs the CLI without a shell.
An account target can now set `cli` (or `target add --cli`) to a name
on PATH or an absolute path. Pi accounts also export
SENPI_CODING_AGENT_DIR and OMO_CODING_AGENT_DIR, which omo reads before
PI_CODING_AGENT_DIR. A missing CLI fails instead of falling back, and a
missing working directory is no longer reported as a missing CLI.
Refs #307
Claude's plugin list also reports other scopes, such as a project-scoped
"(suppressed)@skills-dir" entry for the folder it runs in. The ID check
ran before the scope filter, so one entry the current scope never uses
blocked the whole target - e.g. a claude account target showed
"native plugin list contains an unsupported plugin identifier" when the
UI server ran from the home folder.
A plugin added without choosing an Agent has no binding, so the
dashboard had nowhere to read its version from and always drew the
default icon.
Record the source version on the package when it is added, and let
rows without one ask their source once. Read interface.logo from the
Codex manifest (inside the plugin, image types, 256KB max) and draw
it in the list and in the add preview, falling back to the default
icon.
Since Go 1.23, os.Lstat reports NTFS junctions as irregular files instead
of symlinks (winsymlink, active for go 1.25 in go.mod). Every check that
looked only at os.ModeSymlink therefore missed the junctions skillshare
itself creates on Windows: linked skills could be treated as local folders,
backed up, copied through, or left behind on prune.
Detect links through utils.IsLinkMode, which also asks the OS about the
reparse tag of irregular entries, and count only real symlinks and
junctions as links so OneDrive placeholders and deduplicated files stay
regular files. restore.go also used os.Stat for its link check, which
follows the link and never matched.
Refs #299
Git opens every HTTPS failure with "unable to access", a bad certificate
or a 403 included, and a missing git-lfs says "command not found". Those
were reported as a network problem or a missing source and sent the user
to the wrong fix; they now fall back to the general failure.
Discovery of a managed plugin is also cached per plugin, since two
plugins of one source can have snapshots at different commits.
Opening a plugin cloned its whole source again to learn which Agents
can take it, which took tens of seconds per row. A managed plugin
already has a reviewed snapshot; it answers when its digest still
matches, and the network is used only otherwise.
A failed clone said only that git failed. Git's output is sorted into
an unreachable source or a missing one without passing it on, since it
can hold credentials. The dashboard translates those, notes when a read
or an Agent's CLI is slow, and says when the browser is offline.
When an Agent's plugin inventory failed (for example OpenCode with both
opencode.json and opencode.jsonc present), the inventory call overwrote
the host's installed list with nil, which the API encoded as null. The
Plugins page read its length unguarded, so one misconfigured Agent
crashed the whole page.
Host.fail now restores an empty list, covering every inventory path
that reports through it, and the dashboard reads a null list as empty
so a single Agent can no longer take the page down.
Refs #296
A target declared with `agent` and `config_dir` is another config directory of a
built-in Agent, such as a second Claude account. Its skills and agents paths are
derived from the Agent's defaults at load time, so the config only records the
directory. The same target name works for MCP (written in the Agent's format into
the account's own file), for plugins (the Agent CLI runs with CLAUDE_CONFIG_DIR,
CODEX_HOME or PI_CODING_AGENT_DIR) and for `mcp import --from`. Removing such a
target warns when MCP config still names it instead of failing later at sync.
The dashboard gains an "another account" mode in the add-target dialog and lists
accounts next to the Agents on the MCP and plugin pages.
Refs #289
The source recorded for binding Agents later was also compared on every
install, so adding a second distribution for another Agent under the
same package name was refused. It is only where later bindings come
from; installs are checked per Agent as before.
Adding a plugin forced at least one target, so keeping a plugin in
Skillshare for later meant installing it somewhere and then deselecting
every Agent by hand.
A package now records its source, so it stays managed with no Agent
bound. Agents are bound later from the same source, installing it as it
is then. Removing the last Agent keeps such a plugin; removing the
plugin itself drops it. In the terminal picker, confirming with nothing
selected records it, while Esc still cancels.
The plugin list also shows the installed version as a tag, and an
update check shows old -> new for plugins whose source has moved on.
yaml.Marshal defaults to 4 spaces, and only the config package forced 2.
Saving a plugin re-encodes the whole config.yaml, so installing one
reindented a user's entire file; skill/agent frontmatter and
audit-rules.yaml were written with 4 spaces too.
Route all YAML writers (config, registry, plugins, MCP, frontmatter,
audit rules) through one utils.MarshalYAML so they cannot drift apart.
Pi installs a package by its path, and a package.json with only a pi
field is a valid Pi package, but the missing name blocked Pi for it.
OpenCode also installs by path. A package.json name now only names the
plugin when no other manifest does.