agent-toggle 0.14.0__tar.gz → 0.15.0__tar.gz

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-toggle
-Version: 0.14.0
+Version: 0.15.0
 Summary: Temporarily disable / re-enable AI-agent skills, agents, commands, rules, plugins and MCP servers across harnesses (Claude Code, Codex, Grok, OpenCode, OpenClaw, Copilot, Vibe, Devin, Antigravity)
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/weskao/agent-toggle
@@ -25,6 +25,7 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: telegram-kit<0.3,>=0.2.2
+Requires-Dist: update-kit<0.2,>=0.1
 Provides-Extra: windows
 Requires-Dist: windows-curses; sys_platform == "win32" and extra == "windows"
 Provides-Extra: telegram
@@ -202,7 +203,7 @@ agent-toggle <command> [args]          # or: python3 agent_toggle.py <command> [
 | `status` | health check: harnesses found, types each supports, parked counts, gitignore, untracked parked items, stale live twins, shared dirs, and state entries whose parked item is gone (a `stale` row with the fix command; read-only, still exit `0`) |
 | `list [type]` | what is currently disabled (`--project <dir>` filters to one project) |
 | `cost [--type T]` | estimated startup tokens per item, biggest first (read-only; `--harness H` filters; `--project <dir>` prices a repo's `.claude/` and `.mcp.json` instead of user scope) |
-| `update` | check PyPI now (ignores the cache and a skipped version) and print the commands that update the CLI, the Claude Code plugin (when installed) and the shims; runs none of them. `/agent-toggle update` in Claude Code does the same; opening the pane toasts a newer release once |
+| `update` (also `--update`) | install the newest release tag (`uv tool install --force --no-sources git+<repo>@<tag>`, or `pipx install --force` for a pipx install), then, with the new version, update the Claude Code plugin (when installed) and the shims. Never asks. Exit `0` updated or already current, `1` the check or install failed (offline, rate limit), `2` a dev install (use `git pull`) or an install it cannot manage (the manual command is printed). Messages go to stderr; `--json` has no JSON for it. The `/agent-toggle` pane command has no `update`: run it in a terminal |
 | `install-shims` | write the skill shim into every installed harness that is on in the settings (an off one is skipped unless named with `--harness`); refuses to overwrite a file it did not write (`--dry-run` shows the plan); reports stale legacy `<dir>-disabled/` lines in a harness-home `.gitignore` (`--gitignore` removes them, `--no-gitignore` skips the check); on a terminal, then asks y/n for each problem `doctor` can fix (see [Fixing](#fixing)) |
 | `disable <type> <name>...` | park one or more items (`--dry-run` shows the plan; `--project <dir>` for a repo's own `.claude/` and `.mcp.json`) |
 | `enable <type> <name>...` | put them back (`--dry-run` shows the plan; `--project <dir>` likewise) |
@@ -487,33 +488,37 @@ invalid value in either is ignored with one warning on stderr.
 - `default_harness` is what `disable` / `enable` act on without `--harness`, even when that
   harness is switched off (with one warning on stderr);
   `disable` / `enable --project` without `--harness` still default to `claude`.
-- `language` translates the config menu, help, update prompt and picker chrome; command output stays English.
+- `language` translates the config menu, help, update messages and picker chrome; command output stays English.
 - telegram-kit is a regular dependency now, so reinstalling keeps the Telegram settings working;
   `agent-toggle[telegram]` still resolves.
 
-## Update check
-
-On every invocation (unless disabled) agent-toggle checks PyPI for a newer release. It starts
-before the command and is offered after it, on every exit path; it overlaps the command and waits at most 0.8 s afterwards.
-
-- **On a terminal** (stdin and stderr are TTYs, no `--json`): a panel on stderr, "agent-toggle X
-  is available (you have Y)", with `Update now`, `Skip` and `Skip until next version`, a release
-  notes link, and `↑` `↓` `Enter` `q`. `Update now` runs, in order and stopping at the first
-  failure: `uv tool upgrade agent-toggle` (`pipx upgrade agent-toggle` for a pipx install), then,
-  when the Claude Code plugin is installed, `claude plugin marketplace update agent-toggle` and
-  `claude plugin update agent-toggle@agent-toggle`, then `agent-toggle install-shims`. A failure
-  gives a yellow warning listing the commands to run yourself.
-- **Windows**: a running `agent-toggle.exe` cannot be replaced, so there is no `Update now`; the
-  commands are printed instead (as below), to run once agent-toggle has exited.
-- **Off a terminal** (or `--json`): a stderr line, `agent-toggle X is available (you have Y)`,
-  then those commands, one per line.
-- It never changes the exit code or stdout, so `--json` output stays one document.
-
-It is one `GET https://pypi.org/pypi/agent-toggle/json` on a background thread with a 0.8 s
-timeout, cached for 10 minutes in `~/.agent-toggle/update-check.json`. Nothing is sent beyond a
-normal HTTP request with the User-Agent `agent-toggle-update-check`. See `SECURITY.md`
-for the hardening. To turn it off, set `AGENT_TOGGLE_UPDATE_CHECK=0` or the `update_check` setting
-to off (the settings menu's "Check for updates").
+## Update
+
+`agent-toggle update` (or `agent-toggle --update`) is a self-update through
+update-kit (a small stdlib-only helper library): it asks the repository for its newest `vX.Y.Z`
+tag (`git ls-remote --tags`), and when that is newer than the running version it installs it with
+`uv tool install --force --no-sources git+<repo>@<tag>` (`pipx install --force` for a pipx install).
+It then runs, with the new binary and in order: when the Claude Code plugin is installed,
+`claude plugin marketplace update agent-toggle` and `claude plugin update agent-toggle@agent-toggle`,
+then `agent-toggle install-shims`. A failing step prints a warning and makes the exit code `1`.
+It never prompts and writes to stderr only.
+
+| Exit | Meaning |
+|---|---|
+| `0` | updated, or already up to date |
+| `1` | could not check (offline, rate limit) or the install failed |
+| `2` | a dev install (`git pull` instead), or an install kind it cannot manage (the manual command is printed) |
+
+On Windows a running `agent-toggle.exe` cannot be replaced, so the install runs detached after
+agent-toggle exits (log in `%USERPROFILE%\.update-kit\`).
+
+**The hint.** After a command, on an interactive terminal only (stderr is a TTY, never under
+`--json`, pipes, launchd or timers), agent-toggle prints one stderr line when a newer release
+exists, `agent-toggle vX.Y.Z is available (you have A) — run `agent-toggle update``. The check is
+`git ls-remote` with a 1.5 s timeout, cached for 24 hours (failures too) in
+`~/.update-kit/state.json`. It never changes the exit code or stdout. To turn it off, set
+`AGENT_TOGGLE_UPDATE_CHECK=0` or the `update_check` setting to off (the settings menu's
+"Check for updates").
 
 ## Help
 
@@ -905,7 +910,6 @@ bookkeeping.
 | `parked/<sha8>/` | items parked by `--project` (`<sha8>` = first 8 hex of the SHA-1 of the resolved project dir) |
 | `profiles/` | `<name>.json` profiles (dir `0700`, files `0600`) |
 | `config.json` | [settings](#settings--config-menu) and the Telegram chat id (mode `0600`, atomic write, unknown keys kept) |
-| `update-check.json` | the [update check](#update-check)'s 10-minute cache (mode `0600`, atomic write) |
 
 A killed run loses at most the one item in flight. If that was a flag write or a
 dir move (with its companions), its `pending` record (the full entry, a flag's

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/README.md

@@ -169,7 +169,7 @@ agent-toggle <command> [args]          # or: python3 agent_toggle.py <command> [
 | `status` | health check: harnesses found, types each supports, parked counts, gitignore, untracked parked items, stale live twins, shared dirs, and state entries whose parked item is gone (a `stale` row with the fix command; read-only, still exit `0`) |
 | `list [type]` | what is currently disabled (`--project <dir>` filters to one project) |
 | `cost [--type T]` | estimated startup tokens per item, biggest first (read-only; `--harness H` filters; `--project <dir>` prices a repo's `.claude/` and `.mcp.json` instead of user scope) |
-| `update` | check PyPI now (ignores the cache and a skipped version) and print the commands that update the CLI, the Claude Code plugin (when installed) and the shims; runs none of them. `/agent-toggle update` in Claude Code does the same; opening the pane toasts a newer release once |
+| `update` (also `--update`) | install the newest release tag (`uv tool install --force --no-sources git+<repo>@<tag>`, or `pipx install --force` for a pipx install), then, with the new version, update the Claude Code plugin (when installed) and the shims. Never asks. Exit `0` updated or already current, `1` the check or install failed (offline, rate limit), `2` a dev install (use `git pull`) or an install it cannot manage (the manual command is printed). Messages go to stderr; `--json` has no JSON for it. The `/agent-toggle` pane command has no `update`: run it in a terminal |
 | `install-shims` | write the skill shim into every installed harness that is on in the settings (an off one is skipped unless named with `--harness`); refuses to overwrite a file it did not write (`--dry-run` shows the plan); reports stale legacy `<dir>-disabled/` lines in a harness-home `.gitignore` (`--gitignore` removes them, `--no-gitignore` skips the check); on a terminal, then asks y/n for each problem `doctor` can fix (see [Fixing](#fixing)) |
 | `disable <type> <name>...` | park one or more items (`--dry-run` shows the plan; `--project <dir>` for a repo's own `.claude/` and `.mcp.json`) |
 | `enable <type> <name>...` | put them back (`--dry-run` shows the plan; `--project <dir>` likewise) |
@@ -454,33 +454,37 @@ invalid value in either is ignored with one warning on stderr.
 - `default_harness` is what `disable` / `enable` act on without `--harness`, even when that
   harness is switched off (with one warning on stderr);
   `disable` / `enable --project` without `--harness` still default to `claude`.
-- `language` translates the config menu, help, update prompt and picker chrome; command output stays English.
+- `language` translates the config menu, help, update messages and picker chrome; command output stays English.
 - telegram-kit is a regular dependency now, so reinstalling keeps the Telegram settings working;
   `agent-toggle[telegram]` still resolves.
 
-## Update check
-
-On every invocation (unless disabled) agent-toggle checks PyPI for a newer release. It starts
-before the command and is offered after it, on every exit path; it overlaps the command and waits at most 0.8 s afterwards.
-
-- **On a terminal** (stdin and stderr are TTYs, no `--json`): a panel on stderr, "agent-toggle X
-  is available (you have Y)", with `Update now`, `Skip` and `Skip until next version`, a release
-  notes link, and `↑` `↓` `Enter` `q`. `Update now` runs, in order and stopping at the first
-  failure: `uv tool upgrade agent-toggle` (`pipx upgrade agent-toggle` for a pipx install), then,
-  when the Claude Code plugin is installed, `claude plugin marketplace update agent-toggle` and
-  `claude plugin update agent-toggle@agent-toggle`, then `agent-toggle install-shims`. A failure
-  gives a yellow warning listing the commands to run yourself.
-- **Windows**: a running `agent-toggle.exe` cannot be replaced, so there is no `Update now`; the
-  commands are printed instead (as below), to run once agent-toggle has exited.
-- **Off a terminal** (or `--json`): a stderr line, `agent-toggle X is available (you have Y)`,
-  then those commands, one per line.
-- It never changes the exit code or stdout, so `--json` output stays one document.
-
-It is one `GET https://pypi.org/pypi/agent-toggle/json` on a background thread with a 0.8 s
-timeout, cached for 10 minutes in `~/.agent-toggle/update-check.json`. Nothing is sent beyond a
-normal HTTP request with the User-Agent `agent-toggle-update-check`. See `SECURITY.md`
-for the hardening. To turn it off, set `AGENT_TOGGLE_UPDATE_CHECK=0` or the `update_check` setting
-to off (the settings menu's "Check for updates").
+## Update
+
+`agent-toggle update` (or `agent-toggle --update`) is a self-update through
+update-kit (a small stdlib-only helper library): it asks the repository for its newest `vX.Y.Z`
+tag (`git ls-remote --tags`), and when that is newer than the running version it installs it with
+`uv tool install --force --no-sources git+<repo>@<tag>` (`pipx install --force` for a pipx install).
+It then runs, with the new binary and in order: when the Claude Code plugin is installed,
+`claude plugin marketplace update agent-toggle` and `claude plugin update agent-toggle@agent-toggle`,
+then `agent-toggle install-shims`. A failing step prints a warning and makes the exit code `1`.
+It never prompts and writes to stderr only.
+
+| Exit | Meaning |
+|---|---|
+| `0` | updated, or already up to date |
+| `1` | could not check (offline, rate limit) or the install failed |
+| `2` | a dev install (`git pull` instead), or an install kind it cannot manage (the manual command is printed) |
+
+On Windows a running `agent-toggle.exe` cannot be replaced, so the install runs detached after
+agent-toggle exits (log in `%USERPROFILE%\.update-kit\`).
+
+**The hint.** After a command, on an interactive terminal only (stderr is a TTY, never under
+`--json`, pipes, launchd or timers), agent-toggle prints one stderr line when a newer release
+exists, `agent-toggle vX.Y.Z is available (you have A) — run `agent-toggle update``. The check is
+`git ls-remote` with a 1.5 s timeout, cached for 24 hours (failures too) in
+`~/.update-kit/state.json`. It never changes the exit code or stdout. To turn it off, set
+`AGENT_TOGGLE_UPDATE_CHECK=0` or the `update_check` setting to off (the settings menu's
+"Check for updates").
 
 ## Help
 
@@ -872,7 +876,6 @@ bookkeeping.
 | `parked/<sha8>/` | items parked by `--project` (`<sha8>` = first 8 hex of the SHA-1 of the resolved project dir) |
 | `profiles/` | `<name>.json` profiles (dir `0700`, files `0600`) |
 | `config.json` | [settings](#settings--config-menu) and the Telegram chat id (mode `0600`, atomic write, unknown keys kept) |
-| `update-check.json` | the [update check](#update-check)'s 10-minute cache (mode `0600`, atomic write) |
 
 A killed run loses at most the one item in flight. If that was a flag write or a
 dir move (with its companions), its `pending` record (the full entry, a flag's

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle/__init__.py

@@ -1,3 +1,3 @@
 """agent-toggle: temporarily disable / re-enable AI-agent resources."""
 
-__version__ = "0.14.0"
+__version__ = "0.15.0"

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle/cli.py

@@ -36,7 +36,7 @@ Usage:
     failure, 2 usage error, 3 locked, 4 unsupported pair / harness missing.
     Settings: $HOME/.agent-toggle/config.json (see `config`); env overrides
     AGENT_TOGGLE_UPDATE_CHECK / _COLOR / _LANG / _DEFAULT_HARNESS.
-    Update check: on by default, stderr only; AGENT_TOGGLE_UPDATE_CHECK=0 turns it off.
+    Update hint: on by default, a terminal's stderr only; AGENT_TOGGLE_UPDATE_CHECK=0 turns it off.
 """
 from __future__ import annotations
 
@@ -695,7 +695,7 @@ def build_parser() -> argparse.ArgumentParser:
     sub.add_parser("doctor", parents=[common],
                    help="check harness layouts and state against disk; offers y/n fixes")
     sub.add_parser("update", parents=[common],
-                   help="check for a newer release; show the commands that update everything")
+                   help="install the newest release, then refresh the plugin and shims")
     cf = sub.add_parser("config", parents=[common],
                         help="Telegram settings for the CI failure alerts")
     cf.add_argument("action", nargs="?", choices=("test", "sync-ci"),
@@ -723,9 +723,9 @@ def build_parser() -> argparse.ArgumentParser:
 def main(argv: list[str] | None = None) -> int:
     """Run one command and return its exit code (the process exit code).
 
-    Settings apply first; the update check starts before dispatch and is offered in a
-    `finally`, so every way out (a command, --help/--version, a usage error, Locked,
-    CliError) gets it. It writes to stderr only and never changes the exit code."""
+    Settings apply first; the update hint is printed in a `finally`, so every way out
+    (a command, --help/--version, a usage error, Locked, CliError) gets it. It writes
+    to stderr only and never changes the exit code."""
     argv = normalize_argv(sys.argv[1:] if argv is None else argv)
     try:
         language, color = settings.get("language"), settings.get("color")
@@ -734,16 +734,17 @@ def main(argv: list[str] | None = None) -> int:
     i18n.set_language(language)
     color = scan_color(argv, color)     # --color > env > setting > auto
     i = _command_index(argv)
-    # `update` is the check; `completion` may run on every shell start (eval "$(...)")
-    started = None if argv[i:i + 1] in (["update"], ["completion"]) else update_prompt.start()
+    hint = True
     try:
         return _main(argv, color)
     except KeyboardInterrupt:
-        started = None                  # never prompt after the user hit Ctrl-C
+        hint = False                    # never ask anything after the user hit Ctrl-C
         raise
     finally:
-        with contextlib.suppress(KeyboardInterrupt):     # Ctrl-C while waiting on the check
-            update_prompt.offer(started, json_mode="--json" in argv, color_mode=color)
+        # `update` is the update; `completion` may run on every shell start (eval "$(...)")
+        if hint and argv[i:i + 1] not in (["update"], ["completion"]):
+            with contextlib.suppress(KeyboardInterrupt):
+                update_prompt.hint(json_mode="--json" in argv)
 
 
 def _main(argv: list[str], color: str) -> int:
@@ -804,7 +805,7 @@ def _main(argv: list[str], color: str) -> int:
             if fixes and _interactive(out):
                 doctor.offer_fixes(fixes, out)
         elif cmd == "update":
-            update_prompt.cmd_update(out)
+            return update_prompt.cmd_update()
         elif cmd == "config":
             config.cmd_config(args, out)
         elif cmd == "completion":

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle/help.py

@@ -137,12 +137,13 @@ def commands(default_harness: str = "claude") -> dict[str, Cmd]:
             examples=["config", "config test", "config sync-ci --dry-run"]),
         "update": Cmd(
             "update",
-            t("help.update.summary", "check for a newer release; show how to update"),
-            t("help.update.desc", "Ask PyPI now (ignoring the cache and a skipped version) and, "
-                                  "when a newer release exists, print the commands that update "
-                                  "the CLI, the Claude Code plugin (if installed) and the shims. "
-                                  "Runs none of them."),
-            examples=["update", "update --json"]),
+            t("help.update.summary", "install the newest release; refresh the plugin and shims"),
+            t("help.update.desc", "Install the newest release tag (uv tool or pipx), then, with the "
+                                  "new version, update the Claude Code plugin (if installed) and "
+                                  "reinstall the shims. Never asks; `--update` is the same. A dev "
+                                  "install says to git pull (exit 2). Exit 1 when the check or "
+                                  "the install fails."),
+            examples=["update"]),
         "completion": Cmd(
             "completion bash|zsh|fish|powershell",
             t("help.completion.summary", "print the shell completion script"),

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle/locales/zh_tw_cli.py

@@ -27,9 +27,9 @@ CATALOG: dict[str, str] = {
     "help.cost.summary": "每個項目的預估啟動 token 數，由大到小",
     "help.cost.desc": "估算每個項目啟動時載入多少 token（字元數/4，±25%），"
                       "以及暫存項目已省下多少。",
-    "help.update.summary": "檢查新版本；顯示更新方式",
-    "help.update.desc": "立即向 PyPI 查詢（忽略快取與已略過的版本）；有新版本時，印出更新 CLI、"
-                        "Claude Code 外掛（若已安裝）與 shim 的指令。不會執行任何指令。",
+    "help.update.summary": "安裝最新版本，並更新外掛與 shim",
+    "help.update.desc": "安裝最新的發行標籤（uv tool / pipx），完成後以新版更新 Claude Code 外掛（若已安裝）"
+                        "並重新安裝 shim。不會提問；`--update` 等同 `update`。開發版安裝請改用 git pull（結束碼 2）。",
     "help.completion.summary": "印出 shell 自動補完腳本",
     "help.completion.desc": "在 stdout 印出某個 shell 的 Tab 補完腳本；"
                             "存到該 shell 載入補完的位置（見 README「Shell completion」）。",
@@ -85,16 +85,4 @@ CATALOG: dict[str, str] = {
     "help.opt.help": "顯示說明（`help <command>` 看單一指令）",
     "help.arguments": "參數",
     "help.options": "選項",
-    # update prompt
-    "update.available": "agent-toggle %s 已推出（目前版本 %s）",
-    "update.keys": "↑↓ 選擇 · ⏎ 確認 · q/Ctrl+C 略過",
-    "update.keys_ascii": "上/下 選擇，Enter 確認，q/Ctrl+C 略過",
-    "update.choose": "請選擇 1-3（Enter = 1，q/Ctrl+C = 略過）：",
-    "update.now": "立即更新",
-    "update.skip": "略過",
-    "update.skip_detail": "下次執行再問",
-    "update.skip_version": "略過到下一個版本",
-    "update.skip_version_detail": "有更新的版本時再問",
-    "update.notes": "版本說明：%s",
-    "update.failed": "更新未完成 — 請自行執行：%s",
 }

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle/shims/claude.md.tmpl

@@ -17,7 +17,7 @@ agent-toggle <command> --json
 ```
 If `agent-toggle` is not on PATH, ask the user to install it (`uv tool install agent-toggle`); do not guess where a checkout lives.
 
-Commands: `status` (run first), `doctor` (read-only health check), `update` (checks for a newer release and lists the commands that update it; run none yourself, relay them), `list [type]`,
+Commands: `status` (run first), `doctor` (read-only health check), `update` (installs the newest release, then refreshes the plugin and shims; run it only when the user asks to update), `list [type]`,
 `cost [--type T]` (startup token estimates, biggest first), `disable <type> <name>...` (type `all`: every type with that name),
 `enable <type> <name>...`, `enable --all`, `undo` (reverse the last batch),
 `profile save|apply|diff|delete <name>`, `profile rename <old> <new>`, `profile list` (see Profiles), `migrate`; add `--harness claude|codex|grok|opencode|openclaw|copilot|vibe|devin|agy`

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle/shims/generic.md.tmpl

@@ -20,7 +20,7 @@ You are running in `__HARNESS__`: pass `--harness __HARNESS__` for its own items
 (the default is `claude`). Other harnesses: `--harness claude|codex|grok|opencode|openclaw|copilot|vibe|devin|agy`.
 
 Commands: `status` (run first; shows which types this harness supports),
-`doctor` (read-only health check), `update` (checks for a newer release and lists the commands that update it; run none yourself, relay them), `list [type]`, `cost [--type T]` (startup
+`doctor` (read-only health check), `update` (installs the newest release, then refreshes the plugin and shims; run it only when the user asks to update), `list [type]`, `cost [--type T]` (startup
 token estimates, biggest first), `disable <type> <name>...` (type `all`: every type with that name), `enable <type> <name>...`,
 `enable --all`, `undo` (reverse the last batch), `profile save|apply|diff|delete <name>`,
 `profile rename <old> <new>`, `profile list` (see Profiles); add `--dry-run` to preview. `<type>` = `skill` / `agent` /

agent_toggle-0.15.0/agent_toggle/update_prompt.py

@@ -0,0 +1,80 @@
+"""`update` / `--update` and the one-line "newer release" hint, both through update_kit.
+
+    cmd_update()   # install the newest tag, then refresh the plugin and shims; returns the exit code
+    hint()         # after a command: one stderr line when a newer release exists (TTY only)
+
+No prompts: the old arrow-key panel and its skip-version bookkeeping are gone.
+"""
+from __future__ import annotations
+
+import os
+import subprocess
+import sys
+from collections.abc import Sequence
+from functools import cached_property
+
+from . import __version__, i18n, settings
+from .backends import plugin_cli
+from .ui import theme
+
+DIST = "agent-toggle"
+REPO = "weskao/agent-toggle"
+PLUGIN = "agent-toggle@agent-toggle"
+
+
+def after_steps() -> list[list[str]]:
+    """Run with the NEW binary once the install is done: the Claude Code plugin when it is
+    installed, then the shims."""
+    steps: list[list[str]] = []
+    # bare "claude": what a person pastes; _run() resolves it, as claude is often off PATH
+    if plugin_cli.claude_bin() and any(p["id"] == PLUGIN for p in plugin_cli.list_plugins(lambda _msg: None)):
+        steps += [["claude", "plugin", "marketplace", "update", DIST],
+                  ["claude", "plugin", "update", PLUGIN]]
+    return [*steps, [DIST, "install-shims"]]
+
+
+class _LazySteps(Sequence):
+    """after_steps(), computed on first use: update_kit only reads `after` once an install
+    will happen, so a dev refusal / "already up to date" skips `claude plugin list` (0.8 s)."""
+
+    @cached_property
+    def _steps(self) -> list[list[str]]:
+        return after_steps()
+
+    def __len__(self) -> int:
+        return len(self._steps)
+
+    def __getitem__(self, i):
+        return self._steps[i]
+
+
+def _run(cmd, **kw):
+    """update_kit's subprocess runner: their output stays off stdout and no step may prompt,
+    and a step like `install-shims` must not re-check for updates and print the hint."""
+    if cmd[0] == "claude":
+        cmd = [plugin_cli.claude_bin() or "claude", *cmd[1:]]
+    # update_kit passes its own env (no-prompt git vars) for the install step: merge, don't replace
+    kw["env"] = {**kw.get("env", os.environ), "AGENT_TOGGLE_UPDATE_CHECK": "0"}
+    return subprocess.run(cmd, stdout=sys.stderr, stdin=subprocess.DEVNULL, **kw)
+
+
+def cmd_update() -> int:
+    # imported here, like telegram_kit: a checkout run (`python3 agent_toggle.py`) or a stale env lacks it
+    import update_kit
+    return update_kit.update(app=DIST, repo=REPO, dist=DIST, lang=i18n.LANGUAGE,
+                             after=_LazySteps(), out=sys.stderr, run=_run)
+
+
+def hint(*, json_mode: bool = False) -> None:
+    """One stderr line when a newer release exists: only on a terminal, never under --json,
+    off with the `update_check` setting (env AGENT_TOGGLE_UPDATE_CHECK wins). Never raises."""
+    try:
+        import update_kit
+        if json_mode or not settings.get("update_check") or not update_kit.should_hint(sys.stderr):
+            return
+        latest = update_kit.check(repo=REPO, current=__version__)
+        if latest:
+            print(theme.encodable(update_kit.hint(DIST, latest, __version__, i18n.LANGUAGE), sys.stderr),
+                  file=sys.stderr)
+    except Exception:  # noqa: BLE001 - a hint must never fail the command
+        pass

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-toggle
-Version: 0.14.0
+Version: 0.15.0
 Summary: Temporarily disable / re-enable AI-agent skills, agents, commands, rules, plugins and MCP servers across harnesses (Claude Code, Codex, Grok, OpenCode, OpenClaw, Copilot, Vibe, Devin, Antigravity)
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/weskao/agent-toggle
@@ -25,6 +25,7 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: telegram-kit<0.3,>=0.2.2
+Requires-Dist: update-kit<0.2,>=0.1
 Provides-Extra: windows
 Requires-Dist: windows-curses; sys_platform == "win32" and extra == "windows"
 Provides-Extra: telegram
@@ -202,7 +203,7 @@ agent-toggle <command> [args]          # or: python3 agent_toggle.py <command> [
 | `status` | health check: harnesses found, types each supports, parked counts, gitignore, untracked parked items, stale live twins, shared dirs, and state entries whose parked item is gone (a `stale` row with the fix command; read-only, still exit `0`) |
 | `list [type]` | what is currently disabled (`--project <dir>` filters to one project) |
 | `cost [--type T]` | estimated startup tokens per item, biggest first (read-only; `--harness H` filters; `--project <dir>` prices a repo's `.claude/` and `.mcp.json` instead of user scope) |
-| `update` | check PyPI now (ignores the cache and a skipped version) and print the commands that update the CLI, the Claude Code plugin (when installed) and the shims; runs none of them. `/agent-toggle update` in Claude Code does the same; opening the pane toasts a newer release once |
+| `update` (also `--update`) | install the newest release tag (`uv tool install --force --no-sources git+<repo>@<tag>`, or `pipx install --force` for a pipx install), then, with the new version, update the Claude Code plugin (when installed) and the shims. Never asks. Exit `0` updated or already current, `1` the check or install failed (offline, rate limit), `2` a dev install (use `git pull`) or an install it cannot manage (the manual command is printed). Messages go to stderr; `--json` has no JSON for it. The `/agent-toggle` pane command has no `update`: run it in a terminal |
 | `install-shims` | write the skill shim into every installed harness that is on in the settings (an off one is skipped unless named with `--harness`); refuses to overwrite a file it did not write (`--dry-run` shows the plan); reports stale legacy `<dir>-disabled/` lines in a harness-home `.gitignore` (`--gitignore` removes them, `--no-gitignore` skips the check); on a terminal, then asks y/n for each problem `doctor` can fix (see [Fixing](#fixing)) |
 | `disable <type> <name>...` | park one or more items (`--dry-run` shows the plan; `--project <dir>` for a repo's own `.claude/` and `.mcp.json`) |
 | `enable <type> <name>...` | put them back (`--dry-run` shows the plan; `--project <dir>` likewise) |
@@ -487,33 +488,37 @@ invalid value in either is ignored with one warning on stderr.
 - `default_harness` is what `disable` / `enable` act on without `--harness`, even when that
   harness is switched off (with one warning on stderr);
   `disable` / `enable --project` without `--harness` still default to `claude`.
-- `language` translates the config menu, help, update prompt and picker chrome; command output stays English.
+- `language` translates the config menu, help, update messages and picker chrome; command output stays English.
 - telegram-kit is a regular dependency now, so reinstalling keeps the Telegram settings working;
   `agent-toggle[telegram]` still resolves.
 
-## Update check
-
-On every invocation (unless disabled) agent-toggle checks PyPI for a newer release. It starts
-before the command and is offered after it, on every exit path; it overlaps the command and waits at most 0.8 s afterwards.
-
-- **On a terminal** (stdin and stderr are TTYs, no `--json`): a panel on stderr, "agent-toggle X
-  is available (you have Y)", with `Update now`, `Skip` and `Skip until next version`, a release
-  notes link, and `↑` `↓` `Enter` `q`. `Update now` runs, in order and stopping at the first
-  failure: `uv tool upgrade agent-toggle` (`pipx upgrade agent-toggle` for a pipx install), then,
-  when the Claude Code plugin is installed, `claude plugin marketplace update agent-toggle` and
-  `claude plugin update agent-toggle@agent-toggle`, then `agent-toggle install-shims`. A failure
-  gives a yellow warning listing the commands to run yourself.
-- **Windows**: a running `agent-toggle.exe` cannot be replaced, so there is no `Update now`; the
-  commands are printed instead (as below), to run once agent-toggle has exited.
-- **Off a terminal** (or `--json`): a stderr line, `agent-toggle X is available (you have Y)`,
-  then those commands, one per line.
-- It never changes the exit code or stdout, so `--json` output stays one document.
-
-It is one `GET https://pypi.org/pypi/agent-toggle/json` on a background thread with a 0.8 s
-timeout, cached for 10 minutes in `~/.agent-toggle/update-check.json`. Nothing is sent beyond a
-normal HTTP request with the User-Agent `agent-toggle-update-check`. See `SECURITY.md`
-for the hardening. To turn it off, set `AGENT_TOGGLE_UPDATE_CHECK=0` or the `update_check` setting
-to off (the settings menu's "Check for updates").
+## Update
+
+`agent-toggle update` (or `agent-toggle --update`) is a self-update through
+update-kit (a small stdlib-only helper library): it asks the repository for its newest `vX.Y.Z`
+tag (`git ls-remote --tags`), and when that is newer than the running version it installs it with
+`uv tool install --force --no-sources git+<repo>@<tag>` (`pipx install --force` for a pipx install).
+It then runs, with the new binary and in order: when the Claude Code plugin is installed,
+`claude plugin marketplace update agent-toggle` and `claude plugin update agent-toggle@agent-toggle`,
+then `agent-toggle install-shims`. A failing step prints a warning and makes the exit code `1`.
+It never prompts and writes to stderr only.
+
+| Exit | Meaning |
+|---|---|
+| `0` | updated, or already up to date |
+| `1` | could not check (offline, rate limit) or the install failed |
+| `2` | a dev install (`git pull` instead), or an install kind it cannot manage (the manual command is printed) |
+
+On Windows a running `agent-toggle.exe` cannot be replaced, so the install runs detached after
+agent-toggle exits (log in `%USERPROFILE%\.update-kit\`).
+
+**The hint.** After a command, on an interactive terminal only (stderr is a TTY, never under
+`--json`, pipes, launchd or timers), agent-toggle prints one stderr line when a newer release
+exists, `agent-toggle vX.Y.Z is available (you have A) — run `agent-toggle update``. The check is
+`git ls-remote` with a 1.5 s timeout, cached for 24 hours (failures too) in
+`~/.update-kit/state.json`. It never changes the exit code or stdout. To turn it off, set
+`AGENT_TOGGLE_UPDATE_CHECK=0` or the `update_check` setting to off (the settings menu's
+"Check for updates").
 
 ## Help
 
@@ -905,7 +910,6 @@ bookkeeping.
 | `parked/<sha8>/` | items parked by `--project` (`<sha8>` = first 8 hex of the SHA-1 of the resolved project dir) |
 | `profiles/` | `<name>.json` profiles (dir `0700`, files `0600`) |
 | `config.json` | [settings](#settings--config-menu) and the Telegram chat id (mode `0600`, atomic write, unknown keys kept) |
-| `update-check.json` | the [update check](#update-check)'s 10-minute cache (mode `0600`, atomic write) |
 
 A killed run loses at most the one item in flight. If that was a flag write or a
 dir move (with its companions), its `pending` record (the full entry, a flag's

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle.egg-info/SOURCES.txt

@@ -22,7 +22,6 @@ agent_toggle/spinner.py
 agent_toggle/store.py
 agent_toggle/toml_check.py
 agent_toggle/undo.py
-agent_toggle/update_check.py
 agent_toggle/update_prompt.py
 agent_toggle.egg-info/PKG-INFO
 agent_toggle.egg-info/SOURCES.txt
@@ -85,5 +84,4 @@ tests/test_settings.py
 tests/test_store.py
 tests/test_theme.py
 tests/test_undo.py
-tests/test_update_check.py
 tests/test_update_prompt.py

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/agent_toggle.egg-info/requires.txt

@@ -1,4 +1,5 @@
 telegram-kit<0.3,>=0.2.2
+update-kit<0.2,>=0.1
 
 [telegram]
 telegram-kit<0.3,>=0.2.2

{agent_toggle-0.14.0 → agent_toggle-0.15.0}/pyproject.toml

@@ -27,7 +27,7 @@ classifiers = [
 ]
 requires-python = ">=3.10"
 # telegram-kit is core (stdlib-only) so a plain reinstall can't drop the Telegram settings
-dependencies = ["telegram-kit>=0.2.2,<0.3"]
+dependencies = ["telegram-kit>=0.2.2,<0.3", "update-kit>=0.1,<0.2"]
 
 [project.urls]
 Homepage = "https://github.com/weskao/agent-toggle"