fix(codex): prevent sound hook timeouts

Логин Claude.ai подвисал навсегда: claude auth login ждёт код из браузера в stdin, а хук запускал его в фоне с </dev/null, куда код вставить невозможно. Теперь /add-account поднимает логин со stdin из fifo и печатает ссылку, а /add-account <код> отдаёт код, дожидается обмена и сохраняет новый аккаунт по email из токена. Провал логина больше не маскируется под успешное сохранение текущего аккаунта, а фоновые процессы гасятся строго по своим PID вместо широкого pkill -f. Поведение закреплено пятью тестами в tests/test_fixes.sh.

EFFORT_MAPPING.md

@@ -13,12 +13,12 @@
 
 ## Таблица маппинга
 
-| Claude Code `/effort` | Anthropic (`ai-claude`) | DeepSeek V4 (`ai-deepseek`) | Kimi K2.7/K2.6 (`ai-kimi`) | GLM 5.3 (`ai-glm`) | OpenRouter/GPT-5.5 (`ai-openrouter`) | OpenCode Go (`ai-opencode`) |
+| Claude Code `/effort` | Anthropic (`ai-claude`) | DeepSeek V4 (`ai-deepseek`) | Kimi K2.7/K2.6 (`ai-kimi`) | GLM 5.2 (`ai-glm`) | OpenRouter/GPT-5.5 (`ai-openrouter`) | OpenCode Go (`ai-opencode`) |
 |:---:|:---:|:---:|:---:|:---:|:---:|:---:|
-| `low` | нативно `low` | нативно `low` | thinking on | на стороне сервера | без локального маппинга | без локального маппинга |
+| `low` | нативно `low` | провайдер поднимает до `high` | thinking on | на стороне сервера | без локального маппинга | без локального маппинга |
 | `medium` | нативно `medium` | провайдер поднимает до `high` | thinking on | на стороне сервера | без локального маппинга | без локального маппинга |
 | `high` | нативно `high` | нативно `high` | thinking on | на стороне сервера | без локального маппинга | без локального маппинга |
-| `xhigh` | нативно `xhigh` | провайдер поднимает до `high` | thinking on | на стороне сервера | без локального маппинга | без локального маппинга |
+| `xhigh` | нативно `xhigh` | провайдер поднимает до `max` | thinking on | на стороне сервера | без локального маппинга | без локального маппинга |
 | `max` | нативно `max` | нативно `max` | thinking on | на стороне сервера | без локального маппинга | без локального маппинга |
 
 Для OpenRouter текущий лаунчер не делает локального преобразования effort и отправляет
@@ -45,13 +45,12 @@ low -> medium -> high -> xhigh -> max
 ### DeepSeek V4
 
 ```text
-low -> high -> max
+high -> max
 ```
 
-- 3 реальных уровня.
+- 2 реальных уровня.
-- `low` остаётся `low` — реально снижает thinking бюджет.
+- `low` и `medium` фактически поднимаются до `high`.
-- `medium` и `xhigh` фактически поднимаются до `high`.
+- `xhigh` фактически поднимается до `max`.
-- `max` остаётся `max` — максимальный бюджет thinking.
 - Маппинг выполняется на стороне DeepSeek API.
 - Дефолт `ai-deepseek`: `high`.
 
@@ -66,7 +65,7 @@ on / off
 - Градаций глубины reasoning нет.
 - Дефолт `ai-kimi`: `high`.
 
-### GLM 5.3 (Zhipu AI)
+### GLM 5.2 (Zhipu AI)
 
 ```text
 high / max
@@ -100,7 +99,7 @@ Effort не преобразуется локально — thinking budget не
 | Anthropic | Не нужен | - |
 | DeepSeek V4 | DeepSeek API | На стороне сервера |
 | Kimi K2.7/K2.6 | Moonshot API | На стороне сервера |
-| GLM 5.3 | Zhipu AI API | На стороне сервера |
+| GLM 5.2 | Zhipu AI API | На стороне сервера |
 | OpenRouter/GPT-5.5 | Локального маппинга нет | `ai-openrouter` отправляет значение как есть |
 | OpenCode Go | Локального маппинга нет | `ai-opencode` отправляет значение через routatic-proxy |
 | Legacy GPT proxy | helper-маппинг `xhigh` -> `high` | `~/bin/claude-gpt-effort-proxy.py` или `~/.local/bin/claude-gpt-effort-proxy.py` |
@@ -126,4 +125,4 @@ ai-deepseek                  # без флага - использовать ур
 
 - Для повседневной работы: `high` или `xhigh`.
 - `max` effort имеет реальный эффект у Anthropic и DeepSeek; у Kimi и GLM это всё тот же thinking on / маппинг на стороне сервера.
-- `low`/`medium` у Kimi и GLM фактически не снижают reasoning; у DeepSeek `low` теперь даёт реальный уровень `low` и экономит бюджет.
+- `low`/`medium` у DeepSeek, Kimi и GLM фактически не снижают reasoning.

QUICK_START.md

@@ -15,11 +15,6 @@ bash setup.sh
 exec bash
 ```
 
-Обновление всех AI-инструментов одной командой:
-```bash
-./setup.sh update
-```
-
 ## 2. Настроить сеть (Amnezia + ru-bypass)
 
 Нужно один раз на каждой машине. .ru трафик идёт напрямую,

README.md

@@ -56,15 +56,6 @@ bash setup.sh
 exec bash
 ```
 
-### Обновление одной командой
-
-```bash
-./setup.sh update      # или: ./setup.sh --update
-```
-
-То же, что пункт меню `7) Обновить`: `git pull --ff-only` + перегенерация всех
-`ai-*` лаунчеров через `scripts/ai-setup.sh`. Без аргументов открывается меню.
-
 ## Сеть: ru-bypass + kill switch
 
 Полная документация: [`home-configs/network/README.md`](home-configs/network/README.md)
@@ -160,7 +151,7 @@ bash tests/test_network.sh  # полные тесты
 - `ai-gpt` - нативный OpenAI Codex CLI
 - `ai-deepseek` - Claude Code через DeepSeek API
 - `ai-kimi` - Claude Code через официальный Kimi Code API (`api.kimi.com/coding`)
-- `ai-glm` - Claude Code через Zhipu AI GLM 5.3 API (`api.z.ai/api/anthropic`)
+- `ai-glm` - Claude Code через Zhipu AI GLM 5.2 API (`api.z.ai/api/anthropic`)
 - `ai-openrouter` - Claude Code через OpenRouter (GPT-5.5, Grok, Qwen, GLM 5.2 и др.)
 - `ai-opencode` - Claude Code через OpenCode Go + routatic-proxy (DeepSeek, Kimi, Qwen, GLM, MiniMax, MiMo)
 - `ai-gemini` - нативный Antigravity CLI `agy`
@@ -233,9 +224,9 @@ android init
 
 ## Правила агентов
 
-В `home-configs/GLOBAL_RULES.md` - компактные правила о границах автономности, минимальных
+В `home-configs/GLOBAL_RULES.md` - правила Карпати (Think Before Coding, Simplicity First,
-изменениях, пропорциональной проверке, работе с Git и контекстом проекта, а также пользовательские
+Surgical Changes, Goal-Driven Execution) плюс пользовательские правила (отвечать по-русски,
-настройки: русский язык, доступ к Redmine и запрет на commit/staging без явной команды.
+не коммитить без команды, не делать git add без команды и др.).
 
 `scripts/ai-setup.sh` копирует их в native rule-файлы всех CLI. Лаунчеры обновляют их при каждом запуске.
 

home-configs/GLOBAL_RULES.md

@@ -1,92 +1,100 @@
 # Global AI Agent Rules
 
-Use these defaults across projects, then apply any more specific repository instructions. Prefer clear outcomes and proportional safeguards over process for its own sake.
+Behavioral guidelines to reduce common LLM coding mistakes. Merge with project-specific instructions as needed.
 
-## 1. Resolve Intent Without Needless Blocking
+**Tradeoff:** These guidelines bias toward caution over speed. For trivial tasks, use judgment.
 
-- Inspect the available context before asking the user for information.
+## 1. Think Before Coding
-- Ask a focused question only when an ambiguity would materially change the result, create meaningful risk, or cannot be resolved from the repository or task context.
-- Otherwise choose the simplest reasonable interpretation and state only assumptions that materially affect the result.
-- Surface important tradeoffs and recommend a simpler or safer approach when it better serves the user's goal.
 
-## 2. Keep Changes Minimal and Scoped
+**Don't assume. Don't hide confusion. Surface tradeoffs.**
 
-- Implement only the requested behavior and the support required to make it correct.
+Before implementing:
-- Do not add speculative features, single-use abstractions, unrequested configurability, or handling for implausible scenarios.
+- State your assumptions explicitly. If uncertain, ask.
-- Preserve existing public behavior and local conventions unless the task requires changing them.
+- If multiple interpretations exist, present them - don't pick silently.
-- Do not refactor, reformat, or clean up unrelated code. Mention unrelated issues only when they are relevant risks or blockers.
+- If a simpler approach exists, say so. Push back when warranted.
-- Remove imports, variables, functions, files, or comments made obsolete by your own changes; leave pre-existing dead code alone unless asked.
+- If something is unclear, stop. Name what's confusing. Ask.
-- Every changed line should trace to the requested outcome.
 
-## 3. Act Autonomously Within the Request
+## 2. Simplicity First
 
-- For requests to answer, explain, review, diagnose, or plan, inspect the relevant materials and report the result. Do not implement changes unless the request also asks for them.
+**Minimum code that solves the problem. Nothing speculative.**
-- For requests to change, build, or fix, perform safe in-scope local edits and validation without asking for routine confirmation.
-- Ask before destructive or difficult-to-recover actions, external writes or publication, purchases, or a material expansion of scope.
-- Exception for Eltex merge-request reviews: a bare review request or explicit `eltex-review` invocation authorizes the skill's documented GitLab writes, including inline discussions, approval, and its thank-you note. Explicit chat-only wording keeps the review read-only.
-- Preserve unrelated user changes and adapt to a dirty working tree. Never discard or overwrite work you did not create.
 
-## 4. Plan and Verify Proportionally
+- No features beyond what was asked.
+- No abstractions for single-use code.
+- No "flexibility" or "configurability" that wasn't requested.
+- No error handling for impossible scenarios.
+- If you write 200 lines and it could be 50, rewrite it.
 
-- Define observable success criteria before changing code.
+Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.
-- For non-trivial multi-step work, keep a brief plan whose steps include their verification. Skip planning ceremony for simple tasks.
-- For a bug fix, reproduce the failure first when feasible. Add a focused regression test when it provides durable value and the repository has an appropriate test layer.
-- Run the smallest relevant checks first, then broader checks when the change's risk justifies them.
-- Do not claim completion without fresh evidence. Report the commands run, their results, and anything that remains unverified.
 
-## 5. Keep Git Changes Reviewable
+## 3. Surgical Changes
 
-- Never run `git commit` unless the user explicitly asks for a commit.
+**Touch only what you must. Clean up only your own mess.**
-- Do not stage changes unless the user explicitly asks for staging, committing, or another operation that requires it. Do not disturb changes already staged by the user.
-- Leave tracked edits unstaged so they remain visible in `git diff`.
-- Remember that untracked files do not appear in ordinary `git diff`. Show them with `git status --short`, list their paths in the handoff, and use `git diff --no-index -- /dev/null <path>` when a patch view is useful.
-- Before handing off, inspect both `git diff` and `git status --short` and distinguish your changes from unrelated existing work.
 
-## 6. Load Relevant Project Context
+When editing existing code:
+- Don't "improve" adjacent code, comments, or formatting.
+- Don't refactor things that aren't broken.
+- Match existing style, even if you'd do it differently.
+- If you notice unrelated dead code, mention it - don't delete it.
 
-- Before acting, read the applicable agent instruction files and the project documentation directly relevant to the task.
+When your changes create orphans:
-- Follow referenced documents when they are needed to understand requirements, commands, architecture, or validation. Do not scan unrelated Markdown files indiscriminately.
+- Remove imports/variables/functions that YOUR changes made unused.
-- Use the repository's existing tools, conventions, and test commands unless the task requires otherwise.
+- Don't remove pre-existing dead code unless asked.
 
-## 7. Communication Style
+The test: Every changed line should trace directly to the user's request.
 
-- Reply in Russian, use informal "ты", and communicate like a friendly programmer teammate.
+## 4. Goal-Driven Execution
-- Lead with the outcome and keep the explanation concise while preserving evidence, material caveats, and next steps.
-- Keep identifiers, commands, exact error messages, and source quotations in their original language when accuracy matters.
-- Humor, irony, and swearing are welcome when they fit the conversation; clarity and respect come first.
 
-## 8. Reusable Skills
+**Define success criteria. Loop until verified.**
 
-- After completing the current task, propose a focused reusable skill only when the user has repeated a workflow and reuse would clearly help.
+Transform tasks into verifiable goals:
-- Do not interrupt active work to create a skill, and do not create or edit skill files without explicit user approval.
+- "Add validation" -> "Write tests for invalid inputs, then make them pass"
-- If approved skill changes live outside the current repository, report their exact paths and how the user can inspect them.
+- "Fix the bug" -> "Write a test that reproduces it, then make it pass"
+- "Refactor X" -> "Ensure tests pass before and after"
 
-## 9. Redmine (red.eltex.loc)
+For multi-step tasks, state a brief plan:
+```
+1. [Step] -> verify: [check]
+2. [Step] -> verify: [check]
+3. [Step] -> verify: [check]
+```
 
-- When the request or an Eltex branch identifies a Redmine issue, read its specification, status, and comments before implementing. Eltex branches commonly start with the issue number (for example, `485448_keyboard_opening_fix` maps to issue `485448`).
+Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.
-- Use the personal API key stored only in `~/.config/ai-setup/redmine_key` (mode 600):
-  ```bash
-  curl -fsS -H "X-Redmine-API-Key: $(cat ~/.config/ai-setup/redmine_key)" \
-    "https://red.eltex.loc/issues/<N>.json?include=journals"
-  ```
-- Read `issue.description`, `issue.status.name`, and recent non-empty `issue.journals[].notes`. If the key is missing or the request fails, report that instead of guessing the task requirements.
-- Never print, log, or write the API key into repository files.
 
-## 10. GitLab (gitlab.eltex.loc)
+---
 
-- GitLab access uses **per-project** tokens stored in `~/.config/ai-setup/` (mode 600). Each one is a project access token and works only for its own project:
+**These guidelines are working if:** fewer unnecessary changes in diffs, fewer rewrites due to overcomplication, and clarifying questions come before implementation rather than after mistakes.
 
-  | File | Project | id |
+# Global Rules for All AI Agents
-  | :--- | :--- | :--- |
-  | `gitlab_token_superapp` | `voip/vp-apps/superapp-launcher` | 4928 |
-  | `gitlab_token_common_library` | `voip/vp-apps/common-library` | 4825 |
-  | `gitlab_token_smart_home` | `android_apps/smart_home` | 1409 |
 
-- Pick the file matching the project you are working with. A token from another project returns HTTP 404, not 403, so a wrong token looks exactly like a missing project — check the token before concluding the resource does not exist.
+The rules below are mandatory for every interaction and task. They are intentionally placed after the general coding guidelines above, but they have higher priority when a conflict exists.
-  ```bash
+
-  curl -fsS -H "PRIVATE-TOKEN: $(cat ~/.config/ai-setup/gitlab_token_superapp)" \
+1. **Communication style:**
-    "https://gitlab.eltex.loc/api/v4/projects/4928/merge_requests/241"
+   Always reply in Russian, in a friendly peer-to-peer tone, using informal "ты". Appropriate swearing, humor, sarcasm, and irony are allowed and welcome. Communicate like a live programmer teammate, not like a dry robot.
-  ```
+
-- There is no generic `gitlab_token` fallback any more. Tools that accept an explicit token path (for example the `eltex-review` helper) must be given `--token-file <path>`.
+2. **No commits without an explicit request:**
-- `local.properties` in superapp-launcher also holds `gitlab.token` and `redmine.apiKey`; it is gitignored and must stay that way.
+   Never run `git commit` unless the user has directly and unambiguously asked for it. The final commit always remains with the user, or is made strictly by the user's command.
-- Never print, log, or write these tokens into repository files.
+
+3. **Plain git diff visibility:**
+   All changes must remain visible to the user through the standard `git diff` command. Leave modified files in the working directory unstaged. Do not add files to the Git index with `git add` unless the user explicitly asks for staging, committing, or another action that requires staging, because staging hides changes from plain `git diff`.
+
+4. **Project context:**
+   At the start of work, pay close attention to all provided project `.md` files, because they are provided automatically and contain the current repository's context and specifics.
+
+5. **Reusable skills:**
+   When the user repeats the same instruction, output format, correction, or workflow, treat it as a candidate for a reusable skill.
+
+   Do not interrupt active work just to create a skill. Finish the current task first, then briefly propose the skill name, trigger description, and what files or tools it should contain.
+
+   Do not create or edit skill files silently. Create or update a skill only after explicit user approval.
+
+   Keep skills focused: prefer several small skills over one broad skill. A useful skill should have a precise description for when to use it, short instructions for how to act, and reusable tools, templates, references, or examples when they make the result more stable.
+
+   If skill files are outside the current git repository, clearly state the exact paths changed and how the user can inspect them, because those changes are not visible in the project `git diff`.
+
+6. **Redmine (red.eltex.loc) access:**
+   Redmine issues require authentication. To read a task's spec and comments, use the personal API key stored in `~/.config/ai-setup/redmine_key` (chmod 600). Eltex branches usually start with the issue number (e.g. `485448_keyboard_opening_fix` → issue `485448` → https://red.eltex.loc/issues/485448). Fetch the issue with its comments (journals) via:
+   ```bash
+   curl -s -H "X-Redmine-API-Key: $(cat ~/.config/ai-setup/redmine_key)" \
+     "https://red.eltex.loc/issues/<N>.json?include=journals"
+   ```
+   Read `issue.description` for the spec, `issue.journals[].notes` for comments (recent non-empty ones matter most), `issue.status.name` for status. Keep the key only in that file — never write it into files tracked by git.

home-configs/claude/README.md

@@ -6,7 +6,7 @@ The custom statusline script renders **only ASCII-safe characters and ANSI color
 
 - ASCII brackets `[ ]` for git branch and effort level
 - ANSI 256-color escape sequences for model names, effort levels, and rate-limit percentages
-- The `$` symbol for the DeepSeek balance (USD only)
+- The `¥` symbol (U+00A5) for CNY balance in DeepSeek output
 
 It does **not** use Nerd Font glyphs, powerline separators, or emoji.
 

home-configs/claude/statusline-command.sh

@@ -4,12 +4,6 @@ input=$(cat)
 cwd=$(echo "$input" | jq -r '.cwd')
 model=$(echo "$input" | jq -r '.model.display_name // empty')
 model_id=$(echo "$input" | jq -r '.model.id // empty')
-# Суффикс [1m] в отображении не нужен: окно контекста задаётся
-# CLAUDE_CODE_MAX_CONTEXT_TOKENS, а не суффиксом модели
-model="${model/\[1m\]/}"
-model="${model% \[1M\]}"
-model="${model% (1M context)}"
-model_id="${model_id/\[1m\]/}"
 five_pct=$(echo "$input" | jq -r '.rate_limits.five_hour.used_percentage // empty')
 five_reset=$(echo "$input" | jq -r '.rate_limits.five_hour.resets_at // empty')
 week_pct=$(echo "$input" | jq -r '.rate_limits.seven_day.used_percentage // empty')
@@ -17,18 +11,15 @@ week_reset=$(echo "$input" | jq -r '.rate_limits.seven_day.resets_at // empty')
 ctx_pct=$(echo "$input" | jq -r '.context_window.used_percentage // empty')
 
 
-# Разделитель сегментов: жирная серая средняя точка ·
-_sep() { printf " \033[1m\033[38;5;240m·\033[00m "; }
-
 # Цвет effort: насыщенные, глубже чем пастельные (jewel tones)
 _effort_color() {
   case "$1" in
-    low)    printf '\033[38;5;220mlow\033[00m' ;;                              # золотой
+    low)    printf '\033[38;5;220m[low]\033[00m' ;;                           # золотой
-    medium) printf '\033[38;5;50mmedium\033[00m' ;;                            # насыщенный teal
+    medium) printf '\033[38;5;50m[medium]\033[00m' ;;                          # насыщенный teal
-    high)   printf '\033[38;5;38mhigh\033[00m' ;;                              # глубокий голубой
+    high)   printf '\033[38;5;38m[high]\033[00m' ;;                            # глубокий голубой
-    xhigh)  printf '\033[38;5;206mxhigh\033[00m' ;;                            # насыщенный фиолетовый
+    xhigh)  printf '\033[38;5;206m[xhigh]\033[00m' ;;                          # насыщенный фиолетовый
-    max)    printf '\033[38;5;220mm\033[38;5;114ma\033[38;5;50mx\033[38;5;206m\033[00m' ;;  # радуга
+    max)    printf '\033[38;5;210m[\033[38;5;220mm\033[38;5;114ma\033[38;5;50mx\033[38;5;206m]\033[00m' ;;  # радуга
-    *)      printf '\033[38;5;252m%s\033[00m' "$1" ;;
+    *)      printf '\033[38;5;252m[%s]\033[00m' "$1" ;;
   esac
 }
 
@@ -38,7 +29,6 @@ _brand_color() {
     deepseek)   printf '\033[38;5;69m' ;;   # DeepSeek фирменный синий
     claude)     printf '\033[38;5;173m' ;;  # Anthropic оранжевый
     kimi)       printf '\033[38;5;81m' ;;   # Kimi голубой
-    glm)        printf '\033[38;5;223m' ;;  # Z.ai кремовый (бренд-фон z.ai, #FFE4B5)
     openrouter) printf '\033[38;5;135m' ;;  # OpenRouter фиолетовый
     *)          printf '\033[38;5;223m' ;;  # кремовый (fallback)
   esac
@@ -49,7 +39,7 @@ branch=$(git -C "$cwd" --no-optional-locks symbolic-ref --short HEAD 2>/dev/null
 short_cwd="${cwd/#$HOME/\~}"
 printf "\033[38;5;252m%s\033[00m" "$short_cwd"
 
-[ -n "$branch" ] && { _sep; printf "\033[38;5;111m%s\033[00m" "$branch"; }
+[ -n "$branch" ] && printf " \033[38;5;252m[%s]\033[00m" "$branch"
 if [ -n "$model" ]; then
   brand_color=$(_brand_color "${AI_LAUNCHER:-}")
   effort=$(echo "$input" | jq -r ".effort.level // empty")
@@ -94,21 +84,19 @@ if [ -n "$model" ]; then
         fi
       fi
     fi
-    [ -n "$account" ] && { _sep; printf "%s%s\033[00m" "$brand_color" "$account"; }
+    [ -n "$account" ] && printf " %s[%s]\033[00m" "$brand_color" "$account"
   fi
 
   # Аккаунт/ключ Kimi — показываем alias (account1/account2/...).
   if [ "${AI_LAUNCHER:-}" = "kimi" ]; then
     kimi_account=$(cat "$HOME/.config/ai-setup/kimi_keys/current" 2>/dev/null)
-    [ -n "$kimi_account" ] && { _sep; printf "%s%s\033[00m" "$brand_color" "$kimi_account"; }
+    [ -n "$kimi_account" ] && printf " %s[%s]\033[00m" "$brand_color" "$kimi_account"
   fi
   if [ -n "$effort" ]; then
-    _sep
+    printf " %s%s " "$brand_color" "$model"
-    printf "%s%s " "$brand_color" "$model"
     _effort_color "$effort"
   else
-    _sep
+    printf " %s%s\033[00m" "$brand_color" "$model"
-    printf "%s%s\033[00m" "$brand_color" "$model"
   fi
 fi
 
@@ -150,7 +138,7 @@ if [[ "$model_id" == *deepseek* ]]; then
   cache_file="$HOME/.cache/ai-setup/deepseek_balance"
   if [ -f "$cache_file" ]; then
     balance=$(head -1 "$cache_file")
-    [ -n "$balance" ] && { _sep; printf "\033[38;5;78m%s\033[00m" "$balance"; }
+    [ -n "$balance" ] && printf " \033[38;5;78m%s\033[00m" "$balance"
   fi
 
   # Фоновое обновление баланса (не чаще раза в 30 секунд)
@@ -171,12 +159,13 @@ if [[ "$model_id" == *deepseek* ]]; then
 import sys, json
 d = json.load(sys.stdin)
 infos = d.get('balance_infos', [])
+symbols = {'USD': '\$', 'CNY': '\u00a5'}
 parts = []
 for info in infos:
-    if info.get('currency', '').upper() != 'USD':
+    curr = info.get('currency', '')
-        continue
     total = info.get('total_balance', '0')
-    parts.append(f'\${total}')
+    sym = symbols.get(curr, curr)
+    parts.append(f'{sym}{total}')
 if parts:
     print(' '.join(parts))
 " 2>/dev/null)
@@ -193,7 +182,7 @@ elif [ "${AI_LAUNCHER:-}" = "openrouter" ]; then
   cache_file="$HOME/.cache/ai-setup/openrouter_balance"
   if [ -f "$cache_file" ]; then
     balance=$(head -1 "$cache_file")
-    [ -n "$balance" ] && { _sep; printf "\033[38;5;78m%s\033[00m" "$balance"; }
+    [ -n "$balance" ] && printf " \033[38;5;78m%s\033[00m" "$balance"
   fi
 
   # Фоновое обновление баланса (не чаще раза в 30 секунд)
@@ -258,13 +247,13 @@ else
     five_int=$(printf '%.0f' "$five_pct")
     remaining=$(fmt_remaining "$five_reset")
     color=$(pct_color "$five_int")
-    _sep; printf "%s%s:%s%%\033[00m" "$color" "$remaining" "$five_int"
+    printf " %s%s:%s%%\033[00m" "$color" "$remaining" "$five_int"
   fi
   if [ -n "$week_pct" ] && [ -n "$week_reset" ]; then
     week_int=$(printf '%.0f' "$week_pct")
     remaining=$(fmt_remaining "$week_reset")
     color=$(pct_color "$week_int")
-    _sep; printf "%s%s:%s%%\033[00m" "$color" "$remaining" "$week_int"
+    printf " %s%s:%s%%\033[00m" "$color" "$remaining" "$week_int"
   fi
 fi
 
@@ -274,7 +263,7 @@ fi
 if [ -n "$ctx_pct" ]; then
   ctx_int=$(printf '%.0f' "$ctx_pct")
   color=$(pct_color "$ctx_int")
-  _sep; printf "%sctx:%s%%\033[00m" "$color" "$ctx_int"
+  printf " %sctx:%s%%\033[00m" "$color" "$ctx_int"
 
   # Звуковой сигнал при первом достижении 60%
   alert_file="$HOME/.cache/ai-setup/ctx_alert_state"

scripts/ai-setup.sh

@@ -706,24 +706,25 @@ try:
     d = json.load(sys.stdin)
     available = d.get('is_available', False)
     infos = d.get('balance_infos', [])
-    # Показываем только USD (CNY не выводим)
-    infos = [i for i in infos if i.get('currency', '').upper() == 'USD']
     if not infos:
         print('  \033[0;33m[БАЛАНС]\033[0m Нет данных о балансе')
         sys.exit(0)
+    symbols = {'USD': '\$', 'CNY': '¥'}
     cache_parts = []
     for info in infos:
+        curr = info.get('currency', '???')
         total = info.get('total_balance', '0')
         granted = info.get('granted_balance', '0')
         topped_up = info.get('topped_up_balance', '0')
+        sym = symbols.get(curr, curr)
         status = '✅ доступен' if available else '❌ не активен'
-        print(f'  \033[1;36m💰 Баланс DeepSeek:\033[0m \${total} {status}')
+        print(f'  \033[1;36m💰 Баланс DeepSeek:\033[0m {total} {curr} {status}')
         if float(granted) > 0:
-            print(f'     └─ Начислено: \${granted}')
+            print(f'     └─ Начислено: {granted} {curr}')
         if float(topped_up) > 0:
-            print(f'     └─ Пополнено: \${topped_up}')
+            print(f'     └─ Пополнено: {topped_up} {curr}')
-        cache_parts.append(f'\${total}')
+        cache_parts.append(f'{sym}{total}')
-    # Cache USD balance for statusline
+    # Cache all currencies with symbols for statusline
     if cache_parts:
         cache_dir = os.path.expanduser('~/.cache/ai-setup')
         os.makedirs(cache_dir, exist_ok=True)
@@ -1242,9 +1243,7 @@ fi
 
 _build_ai_sys_prompt > /dev/null  # сохраняет в ~/.codex/AGENTS.md (codex читает авто)
 _setup_codex_sound_hooks
-exec "$codex_bin" --dangerously-bypass-approvals-and-sandbox \
+exec "$codex_bin" --dangerously-bypass-approvals-and-sandbox "$@"
-  -c 'tui.status_line=["model-with-reasoning","git-branch","context-used","five-hour-limit","current-dir"]' \
-  "$@"
 GPTEOF
 chmod +x "$BIN_DIR/ai-gpt"
 
@@ -1561,7 +1560,7 @@ chmod +x "$BIN_DIR/ai-kimi"
 # === ai-glm ===
 cat > "$BIN_DIR/ai-glm" << 'GLMEOF'
 #!/usr/bin/env bash
-# ai-glm - запуск Claude Code через Zhipu AI GLM 5.3 (Anthropic-compatible API)
+# ai-glm - запуск Claude Code через Zhipu AI GLM 5.2 (Anthropic-compatible API)
 source "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/ai-api-helpers.sh" 2>/dev/null || true
 
 key_file="$HOME/.config/ai-setup/glm_key"
@@ -1581,7 +1580,7 @@ glm_base_url="${GLM_BASE_URL:-https://api.z.ai/api/anthropic}"
 
 if [ -n "$api_key" ]; then
   echo -n "Проверка сохранённого GLM ключа... "
-  _claude_test_api "$glm_base_url/v1/messages" "x-api-key: $api_key" "glm-5.3"
+  _claude_test_api "$glm_base_url/v1/messages" "x-api-key: $api_key" "glm-5.2"
   _handle_api_response "GLM" "$_CLAUDE_TEST_CODE" "$_CLAUDE_TEST_BODY" "Пополните баланс: https://z.ai/manage-apikey/apikey-list"
   ret=$_API_RET
   if [ $ret -eq 401 ]; then
@@ -1622,7 +1621,7 @@ if [ -z "$api_key" ]; then
   [ -z "$api_key" ] && { echo "Выход."; exit 1; }
 
   echo -n "Проверяю ключ... "
-  _claude_test_api "$glm_base_url/v1/messages" "x-api-key: $api_key" "glm-5.3"
+  _claude_test_api "$glm_base_url/v1/messages" "x-api-key: $api_key" "glm-5.2"
   _handle_api_response "GLM" "$_CLAUDE_TEST_CODE" "$_CLAUDE_TEST_BODY" "Пополните баланс: https://z.ai/manage-apikey/apikey-list"
   ret=$_API_RET
   if [ $ret -eq 0 ] || [ $ret -eq 429 ]; then
@@ -1645,37 +1644,31 @@ trap 'rm -f "$_PROMPT_FILE"' EXIT INT TERM
 _build_ai_sys_prompt > "$_PROMPT_FILE"
 export AI_LAUNCHER=glm
 export CLAUDE_CONFIG_DIR="$HOME/.config/ai-setup/cfg/glm"
-# Пикер: все три слота = glm-5.3. availableModels НЕ задаём (он схлопывает пикер).
+# Пикер: Opus/Sonnet = GLM 5.2, Haiku = GLM 5.2[1m]. availableModels НЕ задаём
-# Claude Code навязывает 3 слота; незаданный показал бы чужой Claude, поэтому
+# (он схлопывает пикер). Claude Code навязывает 3 слота; незаданный показал бы
-# opus+sonnet+haiku = glm-5.3. Окно контекста у всех слотов задаёт env
+# чужой Claude, поэтому opus+sonnet = glm-5.2, haiku = glm-5.2[1m].
-# CLAUDE_CODE_MAX_CONTEXT_TOKENS ниже — суффикс [1m] больше не нужен.
 _setup_isolated_config glm opus high ''
 # ai-glm стартует на /effort max. max невалиден в settings.json (там только
 # low/medium/high/xhigh), поэтому задаём через CLI --effort, а не _apply_effort.
 # CLAUDE_CODE_ALWAYS_ENABLE_EFFORT заставляет Claude Code отправлять effort в API
-# для GLM (иначе glm-5.3 не распознан как effort-capable и параметр не уходит).
+# для GLM (иначе glm-5.2 не распознан как effort-capable и параметр не уходит).
 # /effort внутри сессии работает; ручной override запуска: AI_EFFORT=high ai-glm.
-# CLAUDE_CODE_DISABLE_1M_CONTEXT НЕ выставляем: GLM 5.3 реально держит 1M контекста,
+# CLAUDE_CODE_DISABLE_1M_CONTEXT НЕ выставляем: GLM 5.2 реально держит 1M контекста,
 # а флаг заставлял бы счётчик ctx: считать окно = 200K (врал в ~5 раз). Прочие
 # лаунчеры (deepseek/kimi/openrouter/opencode) флаг оставляют — у них другой контекст
 # и флаг убирает [1M]-дубли из пикера; ai-claude его и так не использует.
-# CLAUDE_CODE_MAX_CONTEXT_TOKENS: glm-* нет в реестре моделей Claude Code, поэтому
-# окно по умолчанию считается = 200K — ctx: врал бы в ~5 раз, а автокомпакт жал бы
-# сессию в 200K. Задаём реальное окно 1M явно; env применяется к не-claude моделям
-# независимо от суффикса, поэтому [1m] у Haiku-слота не нужен и убран.
 ANTHROPIC_BASE_URL="$glm_base_url" \
 ANTHROPIC_AUTH_TOKEN="$api_key" \
-ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5.3 \
+ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5.2 \
-ANTHROPIC_DEFAULT_OPUS_MODEL_NAME="GLM 5.3" \
+ANTHROPIC_DEFAULT_OPUS_MODEL_NAME="GLM 5.2" \
-ANTHROPIC_DEFAULT_OPUS_MODEL_DESCRIPTION="GLM 5.3 — флагман Zhipu AI" \
+ANTHROPIC_DEFAULT_OPUS_MODEL_DESCRIPTION="GLM 5.2 — флагман Zhipu AI" \
-ANTHROPIC_DEFAULT_SONNET_MODEL=glm-5.3 \
+ANTHROPIC_DEFAULT_SONNET_MODEL=glm-5.2 \
-ANTHROPIC_DEFAULT_SONNET_MODEL_NAME="GLM 5.3" \
+ANTHROPIC_DEFAULT_SONNET_MODEL_NAME="GLM 5.2" \
-ANTHROPIC_DEFAULT_SONNET_MODEL_DESCRIPTION="GLM 5.3 — флагман Zhipu AI" \
+ANTHROPIC_DEFAULT_SONNET_MODEL_DESCRIPTION="GLM 5.2 — флагман Zhipu AI" \
-ANTHROPIC_DEFAULT_HAIKU_MODEL=glm-5.3 \
+ANTHROPIC_DEFAULT_HAIKU_MODEL='glm-5.2[1m]' \
-ANTHROPIC_DEFAULT_HAIKU_MODEL_NAME="GLM 5.3" \
+ANTHROPIC_DEFAULT_HAIKU_MODEL_NAME="GLM 5.2 [1M]" \
-ANTHROPIC_DEFAULT_HAIKU_MODEL_DESCRIPTION="GLM 5.3 — флагман Zhipu AI" \
+ANTHROPIC_DEFAULT_HAIKU_MODEL_DESCRIPTION="GLM 5.2 [1M] — расширенный контекст" \
-CLAUDE_CODE_SUBAGENT_MODEL=glm-5.3 \
+CLAUDE_CODE_SUBAGENT_MODEL=glm-5.2 \
-CLAUDE_CODE_MAX_CONTEXT_TOKENS=1000000 \
 CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 \
 CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 \
 claude --dangerously-skip-permissions --effort "${AI_EFFORT:-max}" --system-prompt-file "$_PROMPT_FILE" "$@"
@@ -2012,7 +2005,7 @@ echo "  На базе Claude Code:"
 echo -e "    ${CYAN}ai-claude${NC}          - Оригинальный Claude Code (Anthropic)"
 echo -e "    ${CYAN}ai-deepseek${NC}        - DeepSeek (через Claude Code, API ключ сохраняется)"
 echo -e "    ${CYAN}ai-kimi${NC}            - Kimi K2.7 Code (через Claude Code, API ключ сохраняется)"
-echo -e "    ${CYAN}ai-glm${NC}             - GLM 5.3 (через Claude Code, Zhipu AI API)"
+echo -e "    ${CYAN}ai-glm${NC}             - GLM 5.2 (через Claude Code, Zhipu AI API)"
 echo -e "    ${CYAN}ai-openrouter${NC}      - OpenRouter (через Claude Code: GPT-5.5, Grok, Qwen, GLM 5.2 и др.)"
 echo -e "    ${CYAN}ai-opencode${NC}        - OpenCode Go (через Claude Code + routatic-proxy: DeepSeek, Kimi, Qwen, GLM, MiniMax, MiMo)"
 echo ""

setup.sh

@@ -18,34 +18,6 @@ mkdir -p "$HOME/.config/ai-setup"
 LOG="$HOME/.config/ai-setup/setup.log"
 _log() { printf '%s [%s] %s\n' "$(date '+%Y-%m-%d %H:%M:%S')" "$1" "$2" >> "$LOG"; }
 
-# Обновление: git pull + перегенерация всех ai-* лаунчеров (пункт меню 7)
-do_update() {
-  local repo_dir
-  repo_dir="$(pwd)"
-  echo -e "${BLD}Обновляем репозиторий...${CLR}"
-  git -C "$repo_dir" pull --ff-only
-  echo ""
-  echo -e "${BLD}Перегенерация скриптов...${CLR}"
-  bash "$repo_dir/scripts/ai-setup.sh"
-}
-
-# CLI-режим: ./setup.sh update (или --update) — обновление без меню
-case "${1:-}" in
-  update|--update)
-    _log "setup" "update (CLI) на $(hostname)"
-    do_update
-    exit $?
-    ;;
-  "")
-    # без аргументов — интерактивное меню ниже
-    ;;
-  *)
-    echo "Неизвестный аргумент: $1"
-    echo "Использование: ./setup.sh [update] — без аргументов открывается меню"
-    exit 1
-    ;;
-esac
-
 if command -v whiptail >/dev/null 2>&1; then
   MENU_ARGS=()
   MENU_ARGS+=("1" "AI-инструменты (установить лаунчеры + ключи)")
@@ -276,7 +248,12 @@ case "$choice" in
     bash tests/test_network.sh
     ;;
   7)
-    do_update
+    REPO_DIR="$(cd "$(dirname "$0")" && pwd)"
+    echo -e "${BLD}Обновляем репозиторий...${CLR}"
+    git -C "$REPO_DIR" pull --ff-only
+    echo ""
+    echo -e "${BLD}Перегенерация скриптов...${CLR}"
+    bash "$REPO_DIR/scripts/ai-setup.sh"
     ;;
   8)
     bash scripts/ai-setup.sh

tests/test_fixes.sh

@@ -41,15 +41,6 @@ test_gpt_no_proxy() {
   fi
 }
 
-# ── ai-gpt: informative Codex status line ───────────────────────────────────
-test_gpt_status_line() {
-  if echo "$GPT_SECTION" | grep -Fq 'tui.status_line=["model-with-reasoning","git-branch","context-used","five-hour-limit","current-dir"]'; then
-    ok "ai-gpt: shows model, branch, context usage, primary limit, and directory"
-  else
-    fail "ai-gpt: missing configured Codex status line"
-  fi
-}
-
 # ── ai-kimi: launches claude with Kimi backend ────────────────────────────
 test_kimi_claude_launcher() {
   if echo "$KIMI_SECTION" | grep -q 'claude --dangerously-skip-permissions' \
@@ -94,13 +85,12 @@ test_glm_claude_launcher() {
 # ── ai-glm: uses official Zhipu AI API ────────────────────────────────────
 test_glm_official_api() {
   if echo "$GLM_SECTION" | grep -q 'api.z.ai/api/anthropic' \
-    && echo "$GLM_SECTION" | grep -q 'ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5.3' \
+    && echo "$GLM_SECTION" | grep -q 'ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5.2' \
-    && echo "$GLM_SECTION" | grep -q 'ANTHROPIC_DEFAULT_HAIKU_MODEL=glm-5.3' \
+    && echo "$GLM_SECTION" | grep -q "ANTHROPIC_DEFAULT_HAIKU_MODEL='glm-5.2\[1m\]'" \
-    && echo "$GLM_SECTION" | grep -q 'CLAUDE_CODE_MAX_CONTEXT_TOKENS=1000000' \
     && echo "$GLM_SECTION" | grep -q 'x-api-key'; then
-    ok "ai-glm: uses official Zhipu AI API (glm-5.3 everywhere, ctx window 1M via env)"
+    ok "ai-glm: uses official Zhipu AI API (glm-5.2 opus/sonnet, glm-5.2[1m] haiku)"
   else
-    fail "ai-glm: must use official Zhipu AI API with glm-5.3, 1M context window and x-api-key"
+    fail "ai-glm: must use official Zhipu AI API with glm-5.2 and x-api-key"
   fi
 }
 
@@ -124,14 +114,20 @@ test_cursor_native_launcher() {
   fi
 }
 
-# ── global rules: source and native rule files ───────────────────────
+# ── global rules: Karpathy-style guidelines and native rule files ───────────
-test_global_rules_source_installed() {
+test_global_rules_include_quality_guidelines() {
-  if [ -s "$GLOBAL_RULES_SOURCE" ] \
+  karpathy_line=$(grep -n '^## 1\. Think Before Coding$' "$GLOBAL_RULES_SOURCE" | head -1 | cut -d: -f1)
+  global_line=$(grep -n '^# Global Rules for All AI Agents$' "$GLOBAL_RULES_SOURCE" | head -1 | cut -d: -f1)
+  if [ -n "$karpathy_line" ] \
+    && [ -n "$global_line" ] \
+    && [ "$karpathy_line" -lt "$global_line" ] \
+    && grep -q 'Always reply in Russian' "$GLOBAL_RULES_SOURCE" \
+    && grep -q 'Plain git diff visibility' "$GLOBAL_RULES_SOURCE" \
     && grep -q 'GLOBAL_RULES_SOURCE=' "$SCRIPT" \
     && grep -q 'cp "$GLOBAL_RULES_SOURCE" "$CONFIG_DIR/global_rules.md"' "$SCRIPT"; then
-    ok "global rules: non-empty source file is installed as global_rules.md"
+    ok "global rules: source file includes English Karpathy guidelines before user rules"
   else
-    fail "global rules: missing source file or installation wiring"
+    fail "global rules: missing source file, English Karpathy guidelines, or user rules"
   fi
 }
 
@@ -295,7 +291,6 @@ test_add_hook_syntax() {
 test_script_syntax
 test_gpt_autoinstall
 test_gpt_no_proxy
-test_gpt_status_line
 test_kimi_claude_launcher
 test_kimi_official_api
 test_kimi_no_artemox
@@ -303,7 +298,7 @@ test_glm_claude_launcher
 test_glm_official_api
 test_gemini_native_launcher
 test_cursor_native_launcher
-test_global_rules_source_installed
+test_global_rules_include_quality_guidelines
 test_native_rule_files_generated
 test_codex_sound_hooks_detach_stdio
 test_fix7_trap_tmp