Commit Graph
13 Commits
Author SHA1 Message Date
陈大猫andCursor 3b3145358d fix(ai): resolve nvmd Windows Codex shims to native codex.exe (#2032)
Settings can probe ~/.nvmd/bin/codex.cmd, but the SDK path needs the real
native binary under versions/<ver>/. Without that override, packaged builds
fall back to optional platform deps we deliberately exclude.

Fixes #2027

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-09 03:33:10 +08:00
陈大猫andCursor 6cb68bf6fc Run agent commands in a subshell so failures cannot close the session (#1945)
* Run agent commands in a subshell so failures cannot close the session

Agent-generated commands were eval'd directly in the user's active login
shell, so a generated script containing `set -e` (or `exit`) terminated
the whole SSH session on the first failure. Wrap the eval in a subshell
unconditionally instead of trying to detect dangerous commands — the
detection approaches from PRs #1852/#1882 grew into shell parsing and
were abandoned. Output stays live in the user's terminal; the trade-off
is that cd/export no longer persist between calls, which the tool
descriptions now tell the model about.

Also fix multi-line command display: synthetic echo now normalizes line
breaks to CRLF (no more staircase indentation in xterm), and the POSIX
wrapper rebuilds newlines via printf so the typed command stays on one
physical line and PS2 continuation echoes ("> ...") no longer leak past
the marker filter.

Fixes #1850

Co-authored-by: Cursor <cursoragent@cursor.com>

* Move shellUtils require to module scope in cattyExecHandlers

The inline require("./ai/shellUtils.cjs") inside the with(ctx) block did
resolve correctly at runtime (bare `require` hits ctx.require, based in
electron/bridges/), but the path only works through that indirection and
reads as wrong relative to this file. Require at module scope with the
unambiguous ../ai path instead.

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-05 23:08:16 +08:00
lengyuqu 66de2db912 Fix CodeBuddy Windows CLI discovery (#1448) 2026-06-12 17:27:52 +08:00
陈大猫 ba784b8b35 fix: resolve .cmd shim to native exe on Windows to avoid spawn EINVAL (#1350) (#1351)
Windows + Node >= 24: spawning .cmd files with shell=false causes EINVAL.
Claude Code v2.1.169 ships as native binary (no cli.js), npm global install
creates only claude.cmd. Netcatty detected claude.cmd but spawned it with
shell:false -> EINVAL.

Changes:
- resolveWindowsShimToNativeExe: new function that reads .cmd/.bat shims
  and resolves to the real .exe using "%~dp0\...\*.exe" pattern matching
- prepareCommandForSpawn: tries native exe resolution first, falls back
  to shell:true wrapping
- resolveClaudeCodeExecutableForSdk: when cli.js not found, looks for
  bin/claude.exe native binary
- 3 new tests for shim resolution and spawn spec
- Codex CLI unaffected (already handles native exe resolution)

Test: 38/38 shellUtils tests pass, npx tsc --noEmit clean
2026-06-09 23:40:21 +08:00
陈大猫andCursor 3662b45121 fix(ai): resolve Windows Codex npm shims before SDK spawn (#1337)
* fix(ai): resolve Windows Codex npm shims before SDK spawn

Codex SDK spawns codexPathOverride without shell:true, so passing
codex.cmd triggers spawn EINVAL on Node 18+. Rewrite npm shims to the
native codex.exe (mirroring #1102 for Claude) on SDK paths only.

Fixes #1101

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(ai): drop unused ctx export and restore test file encoding

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(ai): harden Windows Codex SDK executable resolution

---------

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-06-09 18:34:45 +08:00
陈大猫 3fc56df111 refactor(ai): migrate agent backends from ACP to official SDKs (claude/codex/copilot) (#1229) 2026-06-04 17:11:02 +08:00
陈大猫andClaude Opus 4.7 2a283a4f83 fix(ai): run bundled claude-agent-acp via Electron's Node (#1127)
@zed-industries/claude-agent-acp ships dist/index.js with a
`#!/usr/bin/env node` shebang. We bundle the package and unpack it from
asar, but Windows ignores the shebang entirely and macOS/Linux only
honours it when `node` is on the user's PATH. When `node` was missing,
the resolver fell back to spawning the bare `claude-agent-acp` command,
which only works if the user manually ran
`npm install -g @zed-industries/claude-agent-acp` — see #1118.

Run the bundled script through `process.execPath` with
`ELECTRON_RUN_AS_NODE=1` (matching `resolveMcpServerRuntimeCommand` in
the MCP server bridge) so the embedded Electron acts as the Node
runtime. This makes the bundled copy work with zero external deps on
every supported platform.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-27 21:50:54 +08:00
陈大猫 1aa96c3490 Fix Claude ACP Windows shim launch (#1102) 2026-05-26 14:25:03 +08:00
陈大猫andClaude Opus 4.7 a91fbcdd68 fix #1062: treat SSH shell TMOUT auto-logout as a timeout, not a normal exit (#1067)
* fix #1062: treat SSH shell TMOUT auto-logout as a timeout, not a normal exit

A shell-level TMOUT idle auto-logout makes bash/csh exit cleanly (numeric
exit code, no signal), which is byte-for-byte indistinguishable from a
user-typed `exit` at the SSH protocol level. PR #1057 keyed the
close-vs-keep decision on `streamExited` (numeric code + no signal), so
TMOUT exits were reported as reason "exited" and the tab was auto-closed —
reintroducing the problem from #977.

Verified against a real server that bash TMOUT exits with code 0 / no
signal and prints "timed out waiting for input: auto-logout" to the
channel before it closes. Since exit code/signal can't distinguish it from
an intentional exit, detect that banner in the session's existing rolling
output tail (_promptTrackTail) and report reason "timeout" instead, which
routes to the existing markDisconnected path (keep tab + reconnect). A
normal `exit`/`logout` (no "auto-" prefix) still auto-closes the tab, so
PR #1057's behavior is preserved.

zsh's TMOUT raises SIGALRM (a signal), so it already took the
keep-tab/reconnect path and is unaffected.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Anchor TMOUT auto-logout match to the banner's final line (Codex review)

The detector matched "auto-logout" as an unanchored substring within the
last 256 chars, so command output that merely mentions it (e.g. `grep
auto-logout /etc/profile` while investigating TMOUT) followed by an
intentional `exit` could be misclassified as a timeout and wrongly keep
the tab open. Anchor on the final non-empty line of output instead — the
banner the shell prints right before exiting — which loses no true
positives (verified against the real-server output shape) while rejecting
mid-stream mentions.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-23 10:53:03 +08:00
yuzifu e5a74058ad add test unit 2026-05-06 15:12:17 +08:00
陈大猫 470d9b5aae [codex] Improve ACP agent error diagnostics (#880) 2026-05-01 08:00:50 +08:00
陈大猫andClaude Opus 4.7 d1e6857f76 Drop stale lastIdlePrompt before forcing PowerShell wrapper (#852)
Follow-up to #851 (Codex review comment on 32bab2d4). After that PR,
`resolveEffectiveShellKind` flips an unknown-shell session to PowerShell
based on `session.lastIdlePrompt`, but that field is updated only when
`trackSessionIdlePrompt` recognizes a known prompt shape (default
PowerShell or `user@host[:path][#$]`). On an SSH/Telnet session that
enters PowerShell and then leaves it for a shell with an unrecognized
prompt — cmd.exe (`C:\>`), oh-my-posh / starship / a custom PS1 — the
cached `PS ...>` value persists indefinitely, and every subsequent MCP
command keeps getting wrapped as PowerShell against a non-PowerShell
shell. The new shell errors on the wrapper syntax once per command, and
nothing self-heals until the user reconnects.

Add `getFreshIdlePrompt(session)` which returns the cached prompt only
when the rolling PTY tail (`session._promptTrackTail`) still ends with
it. If the visible last line has moved on — even to a prompt shape we
don't recognize — the cache is treated as expired and downstream
wrapper selection / suffix matching falls back to `shellKind` alone,
which is the correct behavior for the unknown-shell case.

Wire this into the three call sites that previously read
`session.lastIdlePrompt || ""`:
- `aiBridge.cjs:1325` (Catty Agent foreground exec)
- `mcpServerBridge.cjs:1496` (MCP `terminal_execute`)
- `mcpServerBridge.cjs:1584` (MCP `terminal_start` background job)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 15:53:30 +08:00
陈大猫andClaude Opus 4.7 eccb9f2cfc [codex] Fix PowerShell MCP command execution (#851)
* Fix PowerShell MCP command execution

* Harden PowerShell prompt detection and document its scope

- Annotate isPowerShellPrompt and the matching regex in shellUtils with
  a "default prompt only" caveat, so future readers know custom prompt
  themes (oh-my-posh, starship, custom prompt functions) are out of
  scope on purpose, and keep the two regexes in sync.
- Cover edge cases that the original tests left implicit: trailing
  whitespace after the `>`, ANSI-coloured prompts, bare `PS>` with no
  working directory, empty/undefined inputs, and command output that
  merely starts with `PS` (e.g. `PSO>`, `ZIPS>`) so we don't regress
  into mis-wrapping non-PowerShell sessions.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Address multi-agent review findings on PowerShell prompt detection

- Refuse to override an explicit non-PowerShell shellKind. The override
  is only useful when the session has no confirmed shell type (the
  issue #841 case is an SSH session, where shellKind is undefined). On
  a confirmed bash/zsh/fish session a malicious remote process emitting
  a `PS ...>` line could otherwise coerce one mis-wrapped command; this
  closes that foothold while still fixing the original bug.
- Tighten the regex to /^PS(?:\s+\S.*)?>$/ so a literal `"PS >"` line
  is rejected. The default PowerShell prompt never emits that shape, so
  it's a clean spoof signal to ignore.
- Treat `\r` as a line break, not a stripped character, when extracting
  the last idle line. PSReadLine / ConPTY emit bare `\r` to repaint the
  current line; without this, `"PS C:\\old>\rPS C:\\new>"` would match
  as one long doubled prompt that never round-trips through the live
  PTY tail.
- Hoist the regex into shellUtils as `isDefaultPowerShellPromptLine` so
  prompt extraction and wrapper selection share one source of truth.
- Drop a redundant optional-chain on `String.prototype.split().pop()`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Drop dead 'powershell' entry from override set; document shellKind universe

Round-2 review noted that listing "powershell" in
SHELL_KINDS_OPEN_TO_PROMPT_OVERRIDE was a no-op: when the configured
shell kind is already powershell, the override path returns "powershell"
on a match and the fall-through returns "powershell" on a miss, so the
entry only mattered if reverse PS-to-POSIX detection were added later.
Removing it makes the gate's intent ("override only when there's no
confirmed shell type") obvious from the data alone.

Also enumerate the full universe of shellKind values in a comment next
to the set so the next reader doesn't have to grep terminalBridge and
localShell.cjs to know what's excluded and why ("raw" sessions bypass
buildWrappedCommand entirely; "cmd"/"fish" are confirmed and shouldn't
flip to PowerShell on a spoofed remote line).

Add a regression test that locks the current behavior for an explicit
shellKind="powershell" session whose visible prompt looks POSIX (e.g.
nested into WSL/bash) — we keep powershell wrapping. Lock this so a
future maintainer doesn't accidentally introduce reverse detection
without also handling the cross-shell quoting implications.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 15:32:27 +08:00