micro-models-agent 1.5.0 → 1.7.0

package/CHANGELOG.md

@@ -4,6 +4,40 @@ All notable changes to Micro Models Agent (MMA) will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/), and this project adheres to [Semantic Versioning](https://semver.org/).
 
+## [1.7.0] - 2026-09-22
+
+### Added
+
+- **Model capability resolution** (`src/modules/providers/capabilities.ts`, `src/config/types.ts`): a single source of truth for a model's `vision`/`tools` abilities, merged from user override (`config.modelCapabilities[model]`) → live native probe (LM Studio `/api/v1/models` `capabilities`, read by `probeModel` in `src/llm/model-loader.ts`) → provider preset. Exposed to tools via the live `ToolContext.capabilities` getter.
+- **First-class tool image channel** (`src/tools/types.ts` `ToolResult.images`, `src/core/agent/tool-images.ts`): tools can return images; the agent loop queues them onto the next user message (`queueToolImages`), keeping `role:"tool"` content text-only and portable.
+
+### Fixed
+
+- **Vision misdetection for non-`-vl` models** (`src/tools/browser.ts`): the `qwen3.5-9b`-style models report `vision: true` in LM Studio but the name-based `modelSupportsVision` heuristic returned `false`, so screenshots were silently downgraded to a text notice. Name matching is removed; the browser tool now reads `ctx.capabilities.vision`.
+- **Browser screenshots never reached the model** (`src/tools/browser.ts`): screenshots were embedded as base64 text inside the `role:"tool"` output, which backends never decode and the budget truncates. They are now delivered as `image_url` parts; text-only models get the snapshot instead.
+
+### Changed
+
+- **Browser module prompt** (`src/modules/browser/module.ts`): no longer asks the model to self-assess vision support — screenshot delivery is automatic.
+
+## [1.6.0] - 2026-09-22
+
+### Added
+
+- **Tab completion descriptions & navigation** (`src/cli/completer.ts`, `src/ui/line-editor.ts`): completions carry descriptions, file-path completion is registered in the REPL, arrow keys/Tab cycle candidates, Esc restores.
+- **`redactConfigSecrets`** (`src/config/redact.ts`): deep secret redaction for config display.
+
+### Fixed
+
+- **Security-paths wizard answer** (`src/cli/setup.ts`, `src/cli/setup-order.ts`): the question was inverted and a no-op; reworded to restrict semantics and `security.paths.enabled` is now applied.
+- **First-run wizard suppression** (`src/config/config.ts`, `src/cli/main.ts`): `loadConfig` no longer persists default config on first load; one-shot runs warn instead of silently using defaults.
+- **`/wizard` dropped security answers and ignored `provider.entries[]`** (`src/cli/repl-commands.ts`, `src/cli/setup-order.ts`): applies answers centrally and updates the active entry.
+- **`mma config show` leaked secrets** (`src/cli/commands.ts`): secrets redacted by default, `--show-secrets` opts out.
+
+### Changed
+
+- **Chat UI: tool-label accent & separators** (`src/ui/theme.ts`, `src/ui/renderer.ts`, `docs/design/ui/STYLE.md`): the action glyph and the localized tool label now share the category color (`toolAccent`), and a dim `·` separates a status label from its content (reasoning label from reasoning text, tool label from arguments) — the rotating status verb no longer blurs into the model's own words. `Ctrl+O` reprint follows the same accent.
+
 ## [1.5.0] - 2026-09-21
 
 ### Added

package/dist/i18n/en.json

@@ -273,7 +273,7 @@
   "cli.config_key": "Config key (e.g., model, provider.baseUrl)",
   "cli.set_value": "Set a config value",
   "cli.set_done": "Set {key}={value}",
-  "cli.show_config": "Show current config",
+  "cli.show_config": "Show the effective configuration (secrets redacted; use --show-secrets to reveal)",
   "cli.migrate_config": "Migrate legacy config.json to domain format",
   "cli.manage_models": "Manage models",
   "cli.list_models": "List available models",
@@ -348,6 +348,7 @@
   "cli.show_details": "Show session details",
   "cli.delete_session": "Delete a session",
   "cli.first_run": "First run detected. Running setup wizard...",
+  "cli.first_run_oneshot": "No configuration found — running with defaults. Run `mma` (no arguments) once to set up the wizard.",
   "cli.setup_saved_restart": "Settings saved. Restarting with the new configuration...",
   "cli.unknown_cmd": "Unknown command: {name}.",
   "cli.unknown_command": "Unknown command \"{input}\". Did you mean \"{suggestion}\"?",
@@ -558,7 +559,7 @@
   "setup.security_configure": "  Configure security? (y/N)",
   "setup.security_bash_block": "  Block dangerous commands (rm, sudo, ssh, etc.)? (y/N)",
   "setup.security_flags_block": "  Block dangerous flags (--force, -rf, etc.)? (y/N)",
-  "setup.security_paths_deny": "  Allow access to .git/, .env, node_modules/? (y/N)",
+  "setup.security_paths_restrict": "  Restrict access to .git/, .env, node_modules/? (Y/n)",
   "skill.already_loaded": "Skill \"{name}\" is already loaded",
   "skill.not_found": "Skill \"{name}\" not found in any source directory",
   "skill.no_match": "No skills match for: {task}",

package/dist/i18n/ru.json

@@ -271,7 +271,7 @@
   "cli.config_key": "Ключ конфига (напр., model, provider.baseUrl)",
   "cli.set_value": "Установить значение",
   "cli.set_done": "Установлено {key}={value}",
-  "cli.show_config": "Показать текущий конфиг",
+  "cli.show_config": "Показать действующую конфигурацию (секреты скрыты; используйте --show-secrets)",
   "cli.migrate_config": "Мигрировать legacy config.json в domain-формат",
   "cli.manage_models": "Управление моделями",
   "cli.list_models": "Список доступных моделей",
@@ -342,6 +342,7 @@
   "cli.show_details": "Показать детали сессии",
   "cli.delete_session": "Удалить сессию",
   "cli.first_run": "Первый запуск. Запускаем мастер настройки...",
+  "cli.first_run_oneshot": "Конфигурация не найдена — запуск с настройками по умолчанию. Запустите `mma` без аргументов один раз для настройки.",
   "cli.setup_saved_restart": "Настройки сохранены. Перезапуск с новой конфигурацией...",
   "cli.unknown_cmd": "Неизвестная команда: {name}.",
   "cli.unknown_command": "Неизвестная команда \"{input}\". Возможно, вы имели в виду \"{suggestion}\"?",
@@ -556,7 +557,7 @@
   "setup.security_configure": "  Настроить безопасность? (y/N)",
   "setup.security_bash_block": "  Блокировать опасные команды (rm, sudo, ssh и т.д.)? (y/N)",
   "setup.security_flags_block": "  Блокировать опасные флаги (--force, -rf и т.д.)? (y/N)",
-  "setup.security_paths_deny": "  Разрешить доступ к .git/, .env, node_modules/? (y/N)",
+  "setup.security_paths_restrict": "  Ограничить доступ к .git/, .env, node_modules/? (Y/n)",
   "skill.already_loaded": "Скилл \"{name}\" уже загружен",
   "skill.not_found": "Скилл \"{name}\" не найден ни в одном каталоге",
   "skill.no_match": "Нет скиллов подходящих для: {task}",