Commit Graph
2 Commits
Author SHA1 Message Date
Willie c327a26e30 feat: native hooks, MCP tool policy and Pi built-in MCP (v0.23.0) (#304)
* feat(hooks): manage native hooks across agents

Add global and project hook management for ten coding agents with native
format adapters, explicit ownership, revision-guarded synchronization and
selective backup/restore. Preserve unrelated native configuration and leave
hook execution and trust decisions with each agent.

Expose CLI commands and completions alongside dashboard import, editing,
native preview, target/project views, synchronization and backups. Localize
messages and document the workflows across supported locales.

Validation: devcontainer make check, full dashboard Vitest suite,
TypeScript/Vite build, scoped ESLint, isolated CLI lifecycle and real UI
acceptance. Native agent hook execution was not tested.

* feat(mcp)!: sync Pi only through its built-in MCP

Pi could be synced through three different mechanisms (built-in MCP,
pi-mcp-adapter, or the extension file), chosen per server. Mixing them
inside one scope was an error users hit easily, and the adapter and
extension files drifted from what Pi itself reads. Pi's built-in MCP
now covers what the other modes offered, so every Pi server is written
to ~/.pi/agent/mcp.json or <project>/.pi/mcp.json.

Existing setups migrate on the next sync: entries Skillshare wrote to
mcp-adapter.json are removed through the ownership ledger, extension
entries are rewritten in place, and hand-written entries stay put.
piExtension and piOptionsPrune still load (with a notice) so old
configs keep working, and are dropped on the next save. directTools is
kept but no longer rendered, since only the adapter reads it.

Pruning unchanged fields Skillshare wrote earlier is now always on,
the per-project off switch no longer applies to Pi, and the removed
--pi-extension and --pi-options-prune flags explain the change instead
of failing as unknown. Import still reads mcp-adapter.json read-only so
servers kept only there can be brought over.

BREAKING CHANGE: --pi-extension and --pi-options-prune were removed;
Pi servers are always synced to Pi's built-in MCP file.

* fix(mcp): load Pi switches and adapter options left from before 0.23.0

Phase 1 made Pi sync only through its built-in MCP, which left two
upgrade paths broken or misleading:

- A project entry with disabled: true that selected Pi used to work in
  adapter mode, but now fails every sync with "pi cannot turn off".
  Loading now takes pi (and Pi accounts) out of such an entry with a
  notice; an entry left with no targets stays and turns nothing off, so
  an explicit empty list is now valid for a switch. Saving a new switch
  for pi still fails.
- piOptions keys that only pi-mcp-adapter reads (lifecycle, idleTimeout,
  approveTools, ...) were copied into Pi's built-in mcp.json, where
  nothing reads them. Loading drops them with a notice naming each key,
  and saving new ones is refused. Keys neither Pi nor the adapter
  documents still pass through. includeTools/excludeTools are left for
  the tool policy change, which converts them.

Both changes rewrite the YAML node on load, like the phase 1 legacy
fields, so the next save persists them.

* feat(mcp): portable tool policy per server

Users need one place to say which of a server's tools the model sees,
instead of Pi-only adapter options that Pi's built-in MCP no longer
reads. A server's tools{expose, allow, deny} is rendered per Agent:
Pi gets exposure and an ordered toolExposure, Codex gets
enabled_tools/disabled_tools, Copilot gets its tools list. Parts an
Agent cannot hold (OpenCode and Kilo have no per-server field) are
named in the plan and by mcp check rather than dropped silently.

directTools and the adapter's includeTools/excludeTools are converted
on load and on import, so existing configs keep their behavior; the
--direct-tools flag is removed with a pointer to the new flags.
check --live now reports tool names so the dashboard can suggest them.

Refs #140

* feat(ui): drop the Pi mode, prune switch and Direct tools from MCP

Since 0.23.0 Pi syncs only through its built-in MCP and the API no longer
returns piExtension, so the server dialog showed an empty "Pi MCP mode"
select, links to pi-mcp-adapter and pi-mcp-extension, a prune switch the
backend ignores and adapter-only Direct tools. Toggling Pi also opened a
mode setup that no longer exists, and Pi was still offered as an Agent
that can turn a global server off per project, which the backend refuses.

Pi now shows only tool exposure, its other settings and one link to
Pi's MCP docs. The import dialog still sends piExtension, but only to
pick between Pi's mcp.json and the legacy mcp-adapter.json, which the
listed path already tells apart. Settings the config still carries from
before 0.23.0 are explained once on the MCP page from the source's
notices, worded in the dashboard's language.

* fix(ui): clarify hook takeover and localize MCP actions

Identify global and project hooks in sync previews, show full takeover guidance, and link project conflicts to their Hooks tab. Translate MCP pending actions using the existing locale keys.

* feat(mcp): name unapplied tool policy parts in the render preview

The dashboard's server editor needs to say which Agents cannot follow
part of a tool policy before the server is saved. The plan's notices
only cover saved servers, so each rendered target now carries the gaps
from toolPolicyGaps instead of the UI re-implementing the per-Agent rules.

Refs #140

* feat(ui): edit the portable MCP tool policy in the dashboard

Saving a server from the edit dialog rebuilt it from the form and
dropped its tools policy. The draft now carries tools, and a Tools
section edits exposure, allow and deny for every Agent at once. Tool
names can be loaded on demand from the saved server through the live
check, and each selected Agent that does not apply a part is named from
the backend's render preview.

Pi's other settings flag the adapter-only keys and the exposure fields
that Tools now owns, so the form refuses what the backend refuses with
a message that points to Tools. Rows show a short policy marker, and the
plan's tool policy notices are localized on the MCP page.

Refs #140

* feat(mcp): save the config without retired settings on sync

Notices about settings 0.23.0 retired (piExtension, piOptionsPrune,
adapter-only piOptions, Pi switch entries, directTools/includeTools/
excludeTools) only went away when the user happened to save MCP settings.
Sync never rewrote the source, so most users kept seeing them.

Loading already converts those settings in memory, so a sync now writes
the converted config back (global, per project, external source, CLI and
dashboard), after keeping the previous content in the file history.
Dry runs write nothing, and a failed save is reported after the Agent
files are applied. The plan carries `migrates`, so the dashboard's sync
runs MCP even when no Agent file changes.

The CLI notices are one short line per kind with each server named
once, and a sync prints the file it updated with its backup. The
dashboard no longer shows these notices: sync handles them with nothing
for the user to do, and tool policy gaps stay in the server dialog.

* fix(ui): catch MCP tool policy mistakes before saving

The empty "Other Pi settings" box showed a toolExposure example even
while Tools had a setting. That example is exactly what the dialog
refuses in that state, so it read as a saved value that went unflagged.
The example now leaves toolExposure out while Tools is set.

A Deny list that removes every tool Allow keeps was only refused by the
server, in English with raw field names. The Tools section now says so
in the user's language and disables Save; the server check stays.

* fix(ui): move Pi settings help into info tooltips

The Pi block stacked four lines of help under the exposure select, one
of them in raw config words (entry, toolExposure), which pushed the
fields apart. The help now sits behind keyboard-focusable info icons:
one on a new "Pi settings" heading, which keeps the official docs link
beside it since a tooltip cannot hold a link, and one on the exposure
label with a shorter, plain explanation. The line saying Tools sets the
exposure stays, because it explains why the select is locked.

* docs(mcp): describe Pi built-in only and the tool policy

0.23.0 dropped the Pi modes, the prune switch and directTools, and added a
portable tools policy (#140, #143). The MCP reference still told users to
pick pi-mcp-adapter and pass flags that now fail, and said nothing about
what the first sync does to an old config or which Agents apply which part
of a tool policy.

The reference now covers the tools field and flags with a per-Agent table
linked to each Agent's docs, the dashboard Tools section, Pi's built-in
files and options, and an upgrade section: what the first sync converts,
which adapter entries it removes, the config save with its migrate backup,
and what replaces each removed flag. The how-to, dashboard and backup pages
follow.

* docs(skill): teach the built-in skill the Pi and tool policy changes

Agents using the skill would still pass --pi-extension and --direct-tools,
which fail since 0.23.0, and would not know the tools policy or what a sync
does to a 0.22 Pi config. The MCP reference now matches the CLI; the
description is unchanged and stays within its size limit.

* test(mcp): cover Pi migration and the tool policy in the runbook

The runbook still exercised pi-mcp-extension, pi-mcp-adapter and
--pi-options-prune, all removed in 0.23.0. New steps check Pi built-in
cleanup, the first-sync conversion of a 0.22 config with its migrate
backup and untouched hand-written adapter entries, the removed-flag
messages, and what each Agent receives from one tool policy.

* docs(wiki): record the 0.23.0 MCP Pi and tool policy milestone

Keeps the design decisions and evidence limits of #140 and #143 (the
ledger-driven migration, the migrate backup, per-Agent translation rules)
where later work can find them without reading five phase reports.

* docs(i18n): mirror the Pi built-in and tool policy docs in four locales

The zh-Hant, zh-Hans, ja and ko MCP pages still described the removed Pi
modes and flags, and linked to anchors that no longer exist in English.
They now carry the tool policy, Pi, and upgrade sections with the English
anchors and dashboard labels taken from each locale's UI strings.

* feat(mcp): probe an unsaved server definition for the dashboard

Loading tools in the server dialog could only probe the saved config, so
editing a URL and clicking load tested the old address and surfaced a raw
"connection refused" for a server the user was already fixing. A new server
could not load tools at all until saved.

POST /api/mcp/probe runs the existing live probe on the definition in the
request, resolving env and header references as the saved-server check does,
and never saves or syncs. Failures carry an error kind (connect, timeout,
auth, protocol, command, unknown) so the dashboard can explain them in the
user's language and keep the raw text as detail.

* fix(ui): load MCP tools with the dialog's current settings

The Tools section now probes the connection as the form describes it, so
an edited URL or command is what gets tested and a new server can load its
tools before saving. Probe failures read as short localized messages with
the raw error behind Details instead of the probe's English text.

Loaded names are dropped when the connection fields change, since they may
no longer be this server's tools; editing Allow or Deny keeps them.

* docs(mcp): say Load tools probes the dialog's current settings

The MCP and UI references said the button starts the saved server and that
a new server must be saved first, which is no longer true.

* feat(ui): pick MCP tools from a loaded checklist

Typing exact tool names into allow/deny chip inputs made users guess
names the server already knows. The Tools section now loads the
server's tool list and lets users tick which tools the model gets;
unticking writes the exact name to deny, re-ticking removes it (and
adds it to a non-empty allow list), and rows removed by a deny
wildcard are locked with the rule named in a tooltip.

Wildcard and unlisted rules stay editable as chips in the same box,
and an outcome summary tells users which agents honor the list,
which honor it only partly, and which cannot filter tools at all, so
nobody assumes a filter applies where it does not.

* refactor(mcp): drop tools.expose in favor of Pi's own exposure

tools.expose only ever meant something to Pi, yet it sat in the
cross-Agent tool policy, locked Pi's exposure select whenever a policy
was set, and produced an "expose" gap for every other Agent. Pi's
exposure mode now lives only in piOptions.exposure: it can sit next to
a tool policy (only toolExposure still conflicts, since the policy
writes it), allowed tools inherit it, and pi-mcp-adapter's directTools
migrates to it. The flag and field were never released, so they are
removed without a shim.

The dashboard now describes a policy in plain words ("2 tools
excluded") instead of the old "2 denied" shorthand, shows Pi's
exposure on the server list in words, hints that Enter adds a rule,
and no longer leaves "{}" behind when an exposure is set and then
cleared.

* fix(hooks): make import a takeover and plans say what changes

A hands-on trial showed several rough edges in hooks management:

- Importing one per-event candidate such as claude-stop merged every event
  of the file into the entry, so unrelated hooks became managed.
- An imported hook then conflicted on the next sync as an "identical
  unmanaged hook" until --replace, although importing it was the takeover.
- Narrowing an entry was reported as "remove" of the whole file although
  only one event left it, and plans gave no event-level detail.
- A misspelled event such as Stopp was accepted silently and would never run.
- Rewriting a compact settings.json produced half compact, half indented
  JSON and left an empty hooks key Skillshare itself had added.

A name that matches a candidate now picks only that candidate. Saving an
import (Mutation.adopt) records the identical native registrations as the
entry's own even without sync, so the next sync adopts them in place; edits
made outside Skillshare afterwards still conflict. Changes carry added,
updated and removed events, and "remove" is reserved for an entry that leaves
the file. A per-Agent catalog of documented events, from each Agent's
official hooks reference, drives advisory plan warnings and is served at
GET /api/hooks/catalog. JSON edits keep the file's compact or indented style
and drop a hooks key once Skillshare empties the one it created.

* style(ui): keep clean theme switches and picked cards black and white

The clean theme is black-and-white only, so its on switches and picked
cards use ink instead of the green ok and blue accent colors. The playful
theme keeps its own colors.

* feat(hooks): return full file texts from the hooks preview

The dashboard can only show which Agent files a hooks change touches, not
what changes inside them. POST /api/hooks/preview now also returns files[]:
each native file the plan would change, with its current text and the exact
text sync would write, taken from the same plan apply writes from. Other
plan responses stay unchanged, and the route keeps the localhost-only guard.

* feat(hooks): manage Antigravity and Antigravity CLI hooks

Antigravity and its CLI (agy) are skills targets, but their hooks could
not be managed. Both read the same files, ~/.gemini/config/hooks.json and a
project's .agents/hooks.json, so they are one hooks Agent with
antigravity-cli and agy as aliases.

Antigravity's hooks.json maps hook names to blocks of events rather than
one event map. Each Skillshare hook becomes one block named after it, held
as a single owned element, so ownership, adoption, conflicts and restore
work as for the other Agents. Keeping the block's own name lets an import
take the block over in place; a same-named block Skillshare does not own is
a conflict until it is imported or replaced. Disabled blocks are not offered
for import, and edits keep the file's compact or indented style.

The event catalog lists the documented events (seconds timeouts), import
problems are no longer cleared before validation, and the schema and docs
cover the new Agent.

* fix(hooks): replace an edited registration in place

Acceptance found that --replace on a registration Skillshare owned but
someone edited outside appended a second copy: the edited one stayed in the
file, now unmanaged, next to the replacement. When the element at the
record's last written position is owned by no one, replace now overwrites it
there, for both an edit with --replace and 'hooks sync NAME --replace'.

Writing into a one-line array such as "Stop": [{...}] also mixed one-line
and indented layouts; elements written there now stay on that line. Import
candidate lists print an explanation shared by several rows once instead of
on every row.

* feat(ui): rebuild the hooks page around targets, the catalog and real diffs

Editing a hook meant typing event names from memory, repeating each
target by hand and saving JSON the dashboard could not check, and the
preview only listed actions without showing what would change in a
file. The backend now publishes each target's documented events and the
full before/after text of every file a preview touches, so the UI can
guide the whole flow instead of trusting free text.

- Event fields pick from the catalog with descriptions, and show the
  tool filter and timeout unit only where the target has them.
- An empty target tab offers to copy another target's commands, mapping
  events to their closest equivalent and converting timeout units, and
  says when an event has no counterpart.
- Native JSON gets completion, formatting and linting with the line and
  a suggested event, and never drops invalid text when switching back to
  fields; Cmd/Ctrl+S saves without syncing and Esc asks before
  discarding edits.
- The preview shows one card per file with event-level changes and a
  folded line diff that marks the user's own hooks as untouched.
- Import preloads each target's unmanaged hooks, saves only the checked
  ones with adopt so sync does not duplicate them, and keeps pasting as
  a secondary path.
- Antigravity joins the managed targets, the wording says targets
  throughout, and the remove dialog no longer contradicts itself.

* fix(hooks): keep key order and separators when replacing an element

Replacing a registration in place re-encoded it from a Go map, so its keys
came out sorted, and on a one-line array it was minimized without the
spaces the line used: [{"hooks": [{"type": "command", ...}]}] turned into
[{"hooks":[{"command":...,"type":"command"}]}]. The whole line then showed
as changed and read as a rewrite of the user's style.

A replaced element now keeps the replaced one's key order at every depth,
with new keys at the end, and elements written on a one-line array use the
line's own spacing after colons and commas.

* feat(hooks): stop managing a hook while keeping its entries

Removing a hook from the source always queued the removal of its native
entries on the next sync, so there was no way to hand a synced or adopted
hook back to the user. MCP already has this as `remove --keep-files`.

`hooks remove NAME --keep-files` (and `unmanage` in the API mutation)
removes the entry and forgets its ownership records in that scope, so sync
neither removes nor updates what it wrote; the entries show up as unmanaged
and importable again. It is rejected with --sync, like MCP.

The dashboard remove dialog adds "Stop managing" and now says what each
choice does to the target files, including that "Remove from source only"
still deletes the entries on the next sync. Its backup note is neutral and
the dialog is wider to fit the diff.

* fix(mcp): warn that Pi's built-in MCP needs Pi 0.99.0 and no old extension

The first sync after upgrading moves Pi servers off pi-mcp-adapter and
pi-mcp-extension into Pi's mcp.json. A user still on Pi before 0.99.0
silently loses every server, because older Pi does not read that file, and
Pi's docs say an installed extension that registers /mcp replaces the
built-in MCP. Nothing told them either.

The plan now adds one notice, shown by preview, sync and --json, when it
removes an entry Skillshare wrote to mcp-adapter.json, rewrites an owned
entry in pi-mcp-extension's format (the transport field), or the config
still has settings only the extensions read. The ledger check covers a
config that owns adapter entries without saying piExtension.

* docs(mcp): say what Pi needs before the first 0.23.0 sync

The upgrade section described how the config changes but not the two
things a user must do in Pi: run Pi 0.99.0 or later, and remove
pi-mcp-adapter or pi-mcp-extension, which Pi's docs say replace the
built-in MCP. It now shows the sync warning, lists every dropped adapter
option, and says a disabled entry no longer turns a server off for Pi in
that project. The import paragraph claimed sync never changes
mcp-adapter.json; it removes the entries it wrote there.

* fix(ui): make hook diffs, editors and the hooks rail readable

Supervisor and user trial of the rebuilt hooks page found diffs that
read as "we will touch your hook", an editor too small for real hooks,
and text cut short in the rail.

- Diffs: removed lines red and added lines green through new diff theme
  tokens (the one colour exception in the clean theme), removals before
  additions, lines differing only by a trailing comma or whitespace kept
  as context, long lines wrapped, and the "left as is" marker on its own
  line. An adopt-only card says the file is unchanged.
- Editors: native JSON and code editors expand over the dialog and back
  as the same editor, so lint, completion, format, the error count and
  Cmd/Ctrl+S keep working; Esc closes completion first. The invalid-JSON
  explanation sits above the editor.
- Dialogs are 880px like MCP's.
- Main page: full target paths wrap under the name with no inner scroll
  box, the sync card shows the target file, empty descriptions are
  hidden, the unmanaged note avoids an orphaned character, and the row
  menu uses a hooks-specific "each target" label.
- Import names follow the CLI (<target>-<event>) or a script's basename.

* feat(ui): show MCP plan notices before a sync

The API returned the plan's notices, but no dashboard view showed them, so
dashboard users never saw that the 0.23.0 sync rewrites their config or
that Pi's built-in MCP needs Pi 0.99.0 and no old extension. The MCP sync
dialog and the Sync page now list them, translated through mcp.notice.*
keys; a notice the UI does not know is shown as the server sent it.

* fix(install): default CLI installs to the user-local bin directory

Use ~/.local/bin so ordinary script installs and upgrades do not require administrator access. Preserve INSTALL_DIR overrides and existing installations, verify the installed binary directly, and show PATH setup only when needed.

Add offline installer regressions and align installation guidance across all five documentation locales, including automation PATH setup.

Validation: seven offline installer cases, make check in the devcontainer, all five website locale builds, and AI context checks passed.

* fix(ui): let stop managing succeed and open the source config from the header

Stop managing sent a mutation whose plan differs from the removal preview the
dialog had loaded, because the server forgets the ownership records first, so
the revision never matched and every attempt failed as stale. Both the hooks
and MCP remove dialogs now preview that mutation before saving it; it writes no
target file, so there is no diff to review. Its toast also says the entries
stay in the files instead of "Removed".

The source path line under the lists took space most people never read. A
config.yaml button in the page header now shows the full path on hover and
opens the Config page editor; a separate sources.mcp file has no editor there,
so that case copies the path instead.

The Pi chip on a server row no longer mentions other Pi settings.

* test(server): replay the dashboard's hooks and MCP requests against the handlers

UI tests mock the API with one fixed revision, so a dialog that saves with a
revision from the wrong preview passed every vitest while "Stop managing"
failed in real use. These tests send the exact mutation JSON and preview
revision each dialog uses (remove, stop managing, edit, toggle, takeover,
import/adopt, restore, global and project sync) in global, hooks.projects
or mcp.projects, and project-mode scopes, and pin that hooks refuses an
unmanage saved with the removal preview's revision.

* fix(ui): show comma-only line changes in the hooks diff

A line that only gained or lost a trailing comma was shown as unchanged
with its new text, so the preview hid a real byte change to the file.
Lines still align ignoring the comma, so an added sibling does not shift
unrelated lines, but a differing pair now shows as removed and added,
removals first within each change. The user's own event keeps its
"left as is" note on the new line.

* fix(hooks): name the events a Copilot hook file changes

Copilot keeps each hook in its own file, so its plan changes carried no
events and the preview card showed only the action while every other
Agent listed "+ SessionStart" and the like. The planner now compares the
file's hooks map before and after the write and reports added, updated
and removed events the same way shared files do.

* feat(ui): open config.yaml at the section the header button came from

The config.yaml button on the Hooks and MCP pages opened the editor at
the top, leaving the user to scroll for the part that page manages. It
now links to /config?section=hooks or ?section=mcp, and the Config page
puts the cursor on that top-level key and scrolls it into view once the
file is loaded. A config without the key opens at the top as before.

* fix(ui): say what hooks and MCP dialogs will really do

Trying hooks across eight targets showed several places where the
dashboard hid or misstated the outcome:

- Remove and sync writes every pending change of the scope, but the
  hooks and MCP remove dialogs only named the removed entry; they now
  name the other hooks or servers that go out with it.
- With more than three targets, each remove choice spelled out every
  target; it now gives the count.
- Preview cards shrank to their header in a long list and hid the diff.
- A hook's summary listed SessionStart and sessionStart as separate rows
  with no hint of which target runs which; rows now merge case-only
  differences and show their targets' icons.
- Save and sync used the same toast as a plain save.
- The copy panel promised the closest events even when a target had
  none; it now says those events must be picked by hand.
- The Hooks rail's expandable target rows had no accessible name.

* fix(ui): pair hooks diff lines only at the same indentation

Diff lines were aligned ignoring leading whitespace, so appending an
event paired the file's closing braces with braces inside the new block.
Once differing pairs show as removed and added, that turned one comma
change into a run of removed closing braces. Only the trailing comma is
ignored now, so the diff shows the comma pair and the added lines.

* fix(ui): explain each hook removal choice on its own button

The three consequence lines sat in the dialog body, away from the buttons
they describe, and made the dialog long when a hook reached many targets.
Each explanation now shows as a tooltip on its footer button, on hover and
keyboard focus, and is wired through aria-describedby for screen readers.

Tooltip placed button-anchored tips that did not fit below using the
cursor offset, so a two-line tip covered the buttons; it now flips above
the anchor. It also shows nothing without content, which lets the dialog
keep its buttons mounted while the preview loads.

The CJK strings dropped the space between the target text and the
particle, which read as "8 個目標 的".

* fix(hooks): delete the files and folders a removed hook leaves empty

Syncing a hook to an Agent whose hooks file did not exist, then removing
it, left shells behind: `{}` in Claude, Codex, Gemini, Qwen, Droid and
Antigravity files, `{"version":1}` for Cursor, and empty folders such as
~/.copilot/hooks and ~/.pi/agent/extensions. They look like user
configuration and confuse the next import or review.

The ledger now records the shared files Skillshare created and the
folders it made for its outputs. When removal leaves a created file with
nothing but that skeleton, the plan deletes it and says so, and the file
diff shows the deletion. Files and folders that existed before, settings
or comments the user added, and non-empty folders all stay. Apply checks
the skeleton again, so settings added after the preview keep the file.

* fix(hooks): note Copilot's folder trust and use Pi's current package

Running Copilot CLI against a synced project showed that .github/hooks
is skipped until the folder is in trustedFolders, while COPILOT_HOME
hooks run anyway; the Copilot note was the only command Agent note that
did not mention trust. Pi's extension docs now import ExtensionAPI from
@earendil-works/pi-coding-agent, so the hook template follows them. The
old name still loads because the import is type-only, but editors no
longer resolve its types.

* test(hooks): run synced hooks in real Agent CLIs

Unit tests only prove Skillshare writes the native files we believe are
right. This runbook syncs a marker-writing hook for each Agent, starts
the real CLI once per scope with a throwaway HOME and fake credentials,
and checks the markers, so a format or location mistake shows up as a
missing marker. Antigravity and Cursor check credentials before any hook
runs; the runbook lists how to confirm them by hand.

* fix(hooks): write the hooks section in block style

yaml.v3 encodes an empty mapping as flow `{}`, and entries added to it
inherit that style, so the first hook saved after the section was emptied
landed on one unreadable line in config.yaml. The section is now expanded
before it is written, like the MCP section, which also unfolds entries an
earlier save left on one line.

* fix(ui): let Beautify unfold nested one-line YAML

Beautify kept flow collections as written, so a hooks or MCP section
squeezed onto one line stayed that way and the button reported nothing to
change. It now expands flow collections that nest others, and everything
inside them, while short lists such as `targets: [claude, codex]` stay on
one line. Saving still keeps the user's flow style.

* fix(ui): show hook targets once and light them up per event

Each event line repeated its own row of target icons under the header's
icons, and two rows of logos stacked on the right read as clutter. The
header now shows the targets as one overlapping stack, and hovering an
event line fades the targets it does not run on. Screen readers still hear
each line's targets.

* fix(ui): document the hooks section in the config editor

The field panel labelled every key under hooks as an unknown field, which
suggested the section was a mistake. It now explains hooks, its entries,
bindings, events, code, files and projects like the MCP keys.

* docs(hooks): describe the config editor and Copilot folder trust

The hooks page now writes its section in block style, opens the config
editor at hooks:, and the field panel and Beautify understand the
section; none of that was documented. Running Copilot CLI also showed that
project hooks in .github/hooks load only in a trusted folder, which the
destinations section did not say.

* docs(changelog): add 0.23.0 entry

Records native hooks, the MCP tool policy, the dashboard work around both,
and the Pi built-in MCP breaking changes, so users upgrading from 0.22.2
know what moves on their first sync.

* fix(server): validate hooks when the config editor saves

The config editor kept the hooks section as an opaque node and never ran
the hooks checks, so a malformed binding saved successfully and then broke
the Hooks page and `sync --all` until the YAML was repaired by hand. A
project config could also gain `hooks.projects`, which only the global
config may hold. Saving now rejects both with the same message sync gives.

* fix(hooks): keep an Antigravity block's name when importing it

An Antigravity hook is the named block in hooks.json. Importing one under
another name, through `hooks import --from agy NAME` or a renamed row in the
dashboard, planned a new block and left the original unmanaged, so both
ran. Such a takeover is now a conflict that asks for the block's own name,
and the import dialog keeps that name fixed for Antigravity rows.
2026-10-01 02:24:24 +08:00
Willie 4bcd569e25 feat: MCP connections, plugins and a redesigned dashboard (v0.21.0) (#285)
* feat(mcp): sync portable server configuration across six clients

Add inline and external MCP sources with guided CLI and dashboard setup, native imports, conflict previews, ownership tracking, backups, and restore. Support Claude Code, Codex, Cursor, VS Code, OpenCode, and Grok CLI in global and project scopes.

Format YAML on save, document MCP fields in the config assistant, and keep its layout stable for scalar list items. Include schemas, localized UI copy, documentation, and regression and end-to-end coverage.

* fix(config): keep loading config when the MCP section is invalid

A mistake in the mcp section (an unknown key, a targets value that is
not a list, an invalid server field) made config.Load, LoadProject,
Save, SaveIn and ValidateConfig fail. Every command that reads config,
including skills sync, install, list and the dashboard, stopped working
because of settings it never uses.

MCPConfig now keeps a section that does not parse exactly as written and
records the error instead of returning it, so commands unrelated to MCP
load and save the config without touching that section. MCP commands
still validate their source on every run, and the dashboard config
editor calls the new ValidateMCP, so saving an invalid MCP section there
is still rejected.

Also correct the ProjectSources comment: sources.mcp resolves from the
directory containing config.yaml, not from the project root.

* fix(mcp): protect Agent files during concurrent edits and recovery

A release review of the MCP sync engine found ways it could lock users
out, report false conflicts, or change entries it does not own.

Recovery
- A failed write left pending.json behind, and every later MCP command
  refused to run. Failed writes now remove their journal, and a journal
  whose Agent file changed again is discarded, because ownership hashes
  still flag entries that no longer match.
- Preview no longer writes. It reads the ownership that recovery will
  record, byte for byte, so a preview stays valid across the recovery
  that the next write performs under the lock.

Concurrent Agent writes
- Claude Code rewrites ~/.claude.json on its own, so hashing whole files
  expired previews and failed applies. Revisions and pre-write checks
  now cover only the MCP entries, and apply rereads each file and applies
  the previewed entry edits to its current content, keeping settings the
  Agent wrote after the preview.

Comparison and ownership
- Only fields Skillshare renders are hashed, after dropping defaults
  Agents add or omit (type stdio/http, empty env, args or headers,
  enabled: true) and header name case, so equivalent entries are not
  reported as drift. enabled and disabled are compared, so turning off a
  managed server is a conflict instead of being overwritten.
- An unmanaged entry that already matches the source is reported as
  unchanged but no longer claimed, so removing the server never deletes
  an entry the user wrote. Import adopts it explicitly, and without
  --replace refuses to rewrite a differing entry, such as a literal token
  that import turned into an environment reference.
- Sync keeps Agent-only fields such as timeouts in existing entries.

Import
- Agent-only fields become warnings instead of blocking the import.
- streamable-http is accepted, and enabled/disabled flags are checked
  for every client.
- Values carrying a URL password, such as database DSNs, become
  environment references, and credential-like arguments are flagged.

Writes and state
- Save only renders the definition first, so a server that could never
  sync, such as one without targets, is rejected before it is saved.
- A symlinked config.yaml is written through to its target.
- TOML entries are updated in place, keep CRLF line endings, and repeated
  add/remove cycles no longer grow blank lines.
- JSON edits no longer HTML-escape characters such as &.
- GROK_HOME is honored.
- Backups keep the newest 20 per Agent file, pruned by file name, and
  damaged backup files are skipped instead of failing the list.
- GET /api/mcp no longer fails when an Agent that no server uses has an
  unresolvable path.

* feat(mcp): detect the client format of pasted JSON

Import without --from, the add wizard and the dashboard import API
assumed Claude's mcpServers shape unless a client was named, so VS Code
"servers" and OpenCode "mcp" snippets copied from server docs failed to
import. The client is now chosen by the top-level key, and a bare server
object still reads as Claude's single-entry shape.

TOML stays explicit because Codex and Grok share the format; a .toml
file without --from is still read as Codex.

* refactor(mcp): print CLI results through internal/ui

MCP commands wrote plain fmt output and duplicated statusFromErr, unlike
every other command. Plans, backup IDs and save/apply messages now use
the shared ui helpers, so they get the standard status prefixes and
colors; each change line shows the server name, action and target. The
operation log uses statusFromErr. --json output is unchanged.

* fix(server): serve MCP settings only to localhost or IP hosts

MCP settings include stdio commands that Agents execute, so a web page
must not be able to read or change them. The existing Origin check does
not stop DNS rebinding: once an attacker's domain resolves to 127.0.0.1,
the browser sends that domain as both Origin and Host from a loopback
address, and the dashboard also binds 0.0.0.0 in Docker.

Rebinding always needs a domain name, so every MCP route now also
requires a Host that is localhost or an IP literal, whatever the bind
address. Other dashboard routes are unaffected. Opening the dashboard
through a domain name, including a reverse proxy, therefore returns 403
for MCP settings; the message says to use localhost or an IP address,
because rewriting Host at the proxy would still fail the Origin check
and would reopen the hole.

Remove the unused POST /api/mcp/sync endpoint. Configuring with sync
covers it, and the guard wraps every remaining MCP route.

* fix(sync): make sync --all report MCP results reliably

- "sync -g mcp" and "sync --dry-run mcp" ran skills sync, because mcp
  was only recognized as the first argument. It now matches in any
  position.
- sync --all --json printed plain-text errors when the MCP preflight
  failed, and only an error on conflicts. Preflight errors now use the
  JSON error shape, and conflicts also return the credential-free plan so
  scripts can see which entries conflict.
- When resource sync failed, MCP was skipped without notice. Text output
  now warns that MCP settings were not applied; JSON output omits the mcp
  field, which --all otherwise always includes.
- MCP was applied with the preflight revision, which failed whenever the
  resource sync changed config or state in between. It is now planned
  again under the MCP lock, and conflicts found then still stop the
  write.

* fix(sync): recognize agents and extras after flags

sync only checked the first argument for a resource kind, so
"sync --dry-run agents" and "sync -g extras" silently synchronized
skills instead. Sync flags take no values, so the kind now matches in
any position.

The shared parseKindArg keeps its first-argument rule: uninstall, update
and similar commands accept resource names, and a name such as "agents"
must not be read as a kind.

* feat(ui): redesign the MCP dashboard around a server-by-agent matrix

The MCP page listed servers and planned changes as flat text, which made
it hard to see where each server is written and what needs attention.

- A matrix shows each server's status per Agent, summarizes pending
  changes and conflicts, and can hide synced entries. Agents whose MCP
  file or config directory exists are marked as detected; GET /api/mcp
  now returns the detected clients.
- Setup is a two-step dialog: add a remote URL, paste JSON or TOML, or
  import from an Agent, then preview which Agent files change. Importing
  an entry the Agent already has adopts it, Replace overrides only that
  entry, and a target selection equal to the inherited mcp.targets is
  saved without pinning targets on the server.
- Conflicts explain their cause and offer Import from <agent> or
  Replace. Removal lists the affected Agent files and offers removing
  from the source only or removing and syncing. Backups are grouped by
  day, with a preview before restoring a file.
- Agent icons appear on the MCP and Targets pages, and the universal
  target is marked as a path shared across agents. Brand icons come from
  @lobehub/icons-static-svg, with bundled SVGs for Warp and Zed.
- Segmented controls wrap instead of overflowing narrow containers.
- New copy is translated for every supported locale.

* fix(ui): let Sync all apply MCP after syncing resources

Sync all previewed MCP, synced skills, agents and extras, then applied
MCP with the revision from the first preview. When the resource sync
changed config or state, that revision was stale and MCP failed with
"preview again" after everything else had already been synced.

MCP is now rechecked before and after the resource sync by comparing the
reviewed changes (target, name and action) instead of the revision, and
applied with the fresh revision. A different set of changes still stops
before any MCP file is written.

* fix(ui): keep flow-style YAML collections when saving config

Formatting on save forced every collection into block style, so a
hand-written "targets: [claude, codex]" was expanded into a list each
time the config was saved. Each collection now keeps its flow or block
style, indentation is still normalized and comments are preserved, and
flow collections are not padded with spaces.

* docs(mcp): document ownership, normalization and dashboard host rules

- Explain that matching entries stay unmanaged until imported, so
  removing a server leaves them in place, and that import adopts an
  identical entry while --replace rewrites a differing one.
- Note that Agent defaults and header case are not treated as changes,
  that disabling a managed server is a conflict, and that previews stay
  valid while an Agent rewrites unrelated settings.
- Describe Agent-specific fields, streamable-http, GROK_HOME, backup
  retention, recovery by the next MCP write, and the VS Code limitation
  on environment references in headers.
- Document that MCP settings require opening the dashboard by localhost
  or an IP address, with SSH port forwarding for remote machines.
- Update the built-in skill for all six clients and the adoption rule,
  and the MCP runbook for import refusal, Agent-specific fields and
  pulled changes.

* docs(skill): streamline guidance and refresh resource workflows

Keep the built-in skill entrypoint focused and route agent and MCP details to dedicated references. Correct scope, audit, automation, and recovery guidance against the current CLI.

* feat(ui): redesign the dashboard around two styles and a flatter layout

The dashboard had grown page by page, so the same idea looked different
depending on where you met it and pages held controls with nothing to act on.
This rebuilds it on one token set that drives both the Clean and Playful styles
in light and dark, and reshapes the pages around what a person actually does.

Notable changes:

- MCP: replace the server-by-agent matrix with one row per server. The agents a
  server writes to are chips inside the row, so the list grows downwards as more
  CLIs gain MCP support instead of scrolling sideways.
- Settings: collect General, Backup, Log, Health, Extensions and Files into one
  page. The Files tab keeps a panel beside the editor showing what the field
  under the cursor does, the file's shape, or the unsaved changes.
- Audit rules: the YAML editor gets the same panel, plus a Test tab that runs a
  rule's regex against pasted lines.
- Skills and Agents become separate pages; install, search, updates and trash
  move into them. Old routes redirect.
- Tour: rewrite every step's copy, add MCP and Install steps, and anchor each
  step to a control that is always rendered so no step silently times out.
- Give every button a pointer cursor; the browser default does not say a button
  is clickable and each component class had to opt in by hand.

Docs follow the new Settings tabs and the MCP list.

* chore(devcontainer): move dev servers to uncommon ports

5173, 19420 and 8888 are defaults that other projects on the same
machine already hold, so `make devc` failed to bind and the container
never started. Use 45173, 49420 and 48888 instead, identical inside and
outside the container so every banner, log and doc prints the URL that
actually works.

* feat(ui): link dashboard source counts to their pages

The counts in the playful dashboard's source note named a number but led
nowhere, so reaching Skills or MCP from there meant going back to the
sidebar. Each row is now a link to its own page.

The hover fill uses currentColor because .ss-pinnote.src overrides the ink
tokens for the yellow note.

* feat(ui): build and browse hubs from the dashboard

A hub is a JSON index that lists skills, and until now the web UI could
only consume one someone else had already hosted. Writing the index by
hand was the only way to produce one, and the two halves never met: the
dialog that searched hubs and the page that would have built one shared a
config file but nothing else.

The new /hubs page holds both halves behind two tabs:

  Browse   pick a hub, filter its contents by tag or text, install from it
           without leaving the page
  My hubs  assemble an index from installed skills, validate it, export it

A draft is offered as a source in Browse straight away, read from the
draft API rather than a file, so you can see what a recipient sees before
deciding whether to host anything. Where the index is hosted is stored on
the draft and stripped from the exported document, so a reopened hub
remembers where it was published and the published file stays clean.

Hub management moves out of the install dialog, which keeps only the
search-source dropdown and links here.

Routing switches to createBrowserRouter for useBlocker, which guards the
builder against navigating away from unsaved entries.

* docs(extras): add shared instructions recipe for AGENTS.md across agents

Answers a recurring question about sharing one instruction file between AI
agents: extras already covers it, since each agent keeps its user-level
copy in its own directory and one extra can target all of them.

Naming the source file AGENTS.md rather than something neutral matters —
Codex concatenates AGENTS.md files and has no import syntax, so it only
recognises the file by name. Claude Code and Antigravity read a fixed
filename of their own but resolve @-imports, so a one-line CLAUDE.md or
GEMINI.md picks up the synced file.

Says plainly that this is not agent memory. Agents keep their own
learnings in private stores — a Markdown directory for Claude Code, a
database for Codex, non-file storage for Cursor — so that side is a
migration problem, not something syncing files can solve.

* feat(ui): filter skills and agents by target

The Targets column showed which targets a resource reaches, but there was
no way to narrow the list down to one. The filter reads the sync matrix —
the same source that column renders — so a result always matches the icons
the rows show.

Options and counts are derived from the matrix, so no target is offered
that would filter to nothing, and the control stays hidden until at least
two targets have something. A selected target that disappears on a kind
switch or an uninstall falls back to All rather than emptying the list.

The toolbar had 24px to spare at 1440px, so the new control wrapped the
row. Source moves from a 325px segmented control to a dropdown like its
neighbours, and every filter now sizes to its label instead of a fixed
width that truncated the longer values.

* fix(ui): tighten page headers and reach hubs from skills

PageHeader carried mb-7 while .ss-wrap already separates its children with a
28px flex gap. Margins and flex gaps add up rather than collapse, so every
page inside the wrapper opened with a 56px hole under its title. Move the
default onto .ss-pgh so the wrapper can zero it, keeping the 28px for pages
that lay out their own children.

Hubs is a source you install from and a catalog you publish, not inventory
you own, so it no longer sits in the Library group next to Skills, Agents and
Plugins. The Skills page header already links to it and the page already
carries a back link there. Skills stays highlighted while the hub page is
open. Plugins takes the Package icon its own page identifies with, so the
sidebar no longer shows two identical puzzle pieces.

* fix(skill): stop implement-feature from running --no-tui as a command

The dispatch-order note wrote "not --no-tui" as !`--no-tui`. In a skill file
that is the syntax for running a shell command at load time, so the skill died
with "command not found: --no-tui" before any of it could be read. Spell the
condition out in words instead.

* feat(list): toggle model invocation for a skill with M

The only cheap gesture in the list TUI was E, which hides a skill from every
harness. Wanting a skill installed but not auto-loaded meant leaving the TUI
to edit SKILL.md, so people disabled skills they only meant to make
manual-only. M flips disable-model-invocation in place and the detail panel
shows a "manual only" badge so the state is never a blind toggle.

Unlike E this edits the skill file, which upstream may own: update skips
tracked repos with local changes, and reinstalling a skill drops the edit. M
says so and asks before writing in those cases. The flag is written as a
single-line edit rather than a YAML round-trip, and turning it back off
removes the line, so two presses return the file byte for byte and a tracked
repo is clean again. Agents are left alone; their harnesses do not read the key.

Refs #283

* feat(plugin): manage complete plugins across agent tools

Add CLI, TUI, and web flows for discovery, adoption, sync selection, updates, and removal while preserving native package formats and unrelated installations.

Support Claude, Codex, Cursor, Antigravity desktop, Pi, and OpenCode. Update help, completions, built-in skill, and documentation; keep plugins outside sync --all.

* docs(skill): describe the redesigned dashboard in the style skill

The style skill still documented the pre-redesign hand-drawn system, so
pages written with it came out on legacy tokens and outside the ss-*
classes; the new Plugins page was the first casualty.

The rewrite names tokens and classes and points to components.css for
their values, because copied values are what went stale last time. The
ui command docs described Playful by the old look as well.

* fix(ui): keep Playful ghost buttons free of stray shadows

The hover and disabled rules outranked the ghost variant's
box-shadow: none, so a frameless button grew a lone shadow arc in
Playful. Bulk toolbar and toast buttons shared the problem.

Ghost also gets a soft outline in Playful: beside 2px ink frames a bare
label such as Cancel did not read as a button.

* refactor(ui): lay out the Plugins page like the MCP page

The page was built on legacy tokens and plain markup, so it missed the
Playful frames and drifted from every other page. A per-agent column
layout would not have scaled either: each agent that gains plugin
support would add a column.

Plugins now follow the MCP server list: one row per plugin with its
agents as toggles inside the row, so the list only grows downwards.
Per-agent sync and update move to package-level actions in the row
menu, and the dialogs use the standard header, body and footer.

* feat(ui): give plugins a kind colour and per-agent actions

Plugins borrowed a neutral grey tile because no kind colour existed,
which made them the one resource without an identity on the page.
Magenta sits furthest from the five hues already taken.

Laying plugins out as rows had dropped the per-agent sync and update.
Retrying one agent after a partial failure should not re-run the rest,
so each agent in the expanded row gets its own menu. Update is left out
for a deselected agent, which has nothing installed to update.

* feat(ui): count plugins on the Dashboard

The source summary listed every resource kind except plugins, so the
Dashboard under-reported what a source holds. The Playful source note
grows to fit a fifth row.

MCP now sits above Plugins in the sidebar to match that order, and its
Dashboard label reuses the sidebar's "MCP" instead of a longer string
of its own.

* feat(scripts): add a host verification environment

Plugin and MCP work talks to the real Agent CLIs, and those live on the
host, not in the devcontainer, where every target reports "not installed".
Verifying by hand meant either pointing a build at real config or
assembling a dozen environment variables each time.

`make verify` builds a host binary and runs it against .verify-home/ in
the repo, redirecting HOME, the XDG dirs and each agent's own config
variable. The environment starts from `env -i`, so a stray
SKILLSHARE_CONFIG in the caller's shell cannot pull the run back to real
paths. Living inside the repo is safe because project-mode detection
only inspects the working directory and never walks up to a parent.

It also seeds the dashboard cache from ui/dist. A dev build serves a
placeholder rather than the working tree, because runtime download only
applies to numbered versions, so the UI was previously unverifiable
without a release.

* fix(ui): show progress while a plugin selection is applied

Toggling an agent shells out to that agent's own CLI, which takes
seconds. The row gave no sign of it: the toggle only greyed out, and
the confirmation badge is hidden while a target is deselected, so
turning one on looked like nothing happened at all.

The toggle being applied now lifts out of the greyed state and spins in
its badge. Agent versions also get room for their real strings, since
"2.1.276 (Claude Code)" was truncated to a width chosen for a bare
version number.

verify.sh gains a 'run' subcommand, so the same throwaway environment
can be driven from a script rather than only an interactive shell.

* fix(ui): spin the control that started a plugin action

Reading the inventory shells out to six agent CLIs and takes about two
and a half seconds, and every action refreshes it afterwards. Nothing on
the page said so: a menu action closed the menu and left the page still,
so the wait read as a freeze.

Each control now tracks its own work. Check, sync and import spin in
place, a menu action spins the row it acts on, and the agents refresh
follows the query itself rather than the mutation flag.

* feat(ui): split MCP add and import by where the configuration comes from

"Import from a target" opened a dialog whose second tab pasted a snippet,
which is not a target and usually comes from a provider's setup page rather
than a tool the user already configured. Pasting is a way to add a server,
so it moves under "Add server" beside the form, and importing now means one
thing: read what an installed Agent already has.

The paste side gains a file picker, matching `mcp import --file`; the API
only ever took content, so the browser reads the file and fills the editor.

The target picker no longer waits for a successful parse, and a footer that
said "writes 0 config files" next to a greyed-out button now names what is
missing instead.

* fix(ui): label the plugins result strip as the last action

The strip and the plugin inventory below it used the same list styling with
no heading on either, so two rows carrying the same plugin name read as one
list. The status word made it worse: "imported" is one of the outcomes
lifecycle.go returns for an action, but it sat where an inventory row shows
state, so it looked like a property of the plugin.

Give it the section heading the Agents list already has, plus a count and a
dismiss button, so it reads as the report it is.

* feat(plugins): expand compatibility and clarify setup guidance

Support additional native targets, safe source links, Git refs, and explicit OpenCode entries while exposing per-target capabilities across CLI, TUI, and UI.

Separate unsupported integrations from environment errors, explain next steps in plain language, and link official documentation. Update built-in guidance, website docs, completions, and regression coverage.

* feat(plugin): serve a plugin's files and translatable discovery problems

The dashboard could only show a plugin's name and targets, so reviewing
what a plugin ships meant opening the snapshot by hand. Files and
ReadFile expose the reviewed local copy read-only through os.OpenRoot,
which keeps reads inside the snapshot even across symlinks.

Discovery problems now carry a key and arguments next to the English
sentence, so the dashboard can translate them while the CLI keeps its
wording. Inventory asks every Agent CLI in parallel and Packages answers
from config alone, so the plugin list no longer waits on the slowest CLI.

* feat(mcp): sync to Pi, show native config, and lay out written entries

Pi has no built-in MCP support, so a server targeting it must name the
third-party extension that will read the file. `--pi-extension` and the
`piExtension` source field record that choice, and sync writes the file
format of the chosen package instead of guessing.

RenderNative and POST /api/mcp/render return what each Agent would
receive for a definition, with secrets kept as references, so the
dashboard can show the result before anything is written.

Sync used to insert entries on a single line and never revisit them. A
managed entry that still sits on one line is now reported as an update
and rewritten one field per line; entries formatted by hand are left
alone.

* feat(ui): choose plugin Agents like MCP, browse plugin files, view MCP config

The plugin fold listed bound Agents with a per-Agent menu, while MCP let
you simply tick the Agents you want. Plugins now work the same way: the
fold discovers where the source can also go, shows those Agents as
unticked toggles, and a tick previews an install. The per-Agent menu is
gone, and updating a plugin targets only the Agents that support it
instead of hiding the action when one of them does not.

A plugin with a local copy can be browsed in a read-only files dialog,
using the same tree as a skill's Files tab. The MCP "view config" moved
out of a fold in the server form into its own view, reachable from the
server menu, so the native result has room to be read. The MCP dialogs
also ask for the Pi extension when Pi is selected.

Playful tile names drop from 25px to 22px: "brainstorming" was 1px wider
than the tile and lost its last letter to the next line.

* docs: reshoot dashboard screenshots for the redesigned UI

The README and website still showed the pre-redesign dashboard, which no longer matches what users see after upgrading.

* docs: describe MCP, plugins and the new dashboard for v0.21.0

The README still announced v0.20.0 and never mentioned plugins, and its
MCP section listed four clients. The built-in skill and the docs pick up
the Pi extension setup, the formatting-only MCP update, and the plugin
fold and files dialog, so agents and users are not surprised by them.
2026-09-19 03:14:03 +08:00