feat: add /stop command to cancel running Claude calls

docs/plans/2026-02-28-dynamic-command-menu-design.md

@@ -0,0 +1,215 @@
+# Dynamic Command Menu & Plugin Manager
+
+**Date:** 2026-02-28
+**Status:** Approved
+
+## Summary
+
+Add a `/menu` command (wired to Telegram's persistent menu button) that dynamically discovers all Claude Code skills, commands, and plugins from the filesystem and presents them as a navigable inline keyboard. Includes full plugin management (browse, enable/disable, install/update).
+
+## Requirements
+
+1. **Dynamic discovery** — Scan `~/.claude/` on each menu open (no caching)
+2. **Unified menu** — Bot commands + Claude Code skills/commands + custom skills in one place
+3. **Natural categories** — Plugin names as groupings (not hardcoded taxonomy)
+4. **Plugin management** — Browse, enable/disable, install new, update existing
+5. **Persistent menu button** — Always-visible access via Telegram's menu button
+6. **In-place navigation** — Edit existing message, don't spam new ones
+
+## Architecture
+
+### Navigation Model
+
+```
+/menu (top level)
+├── 🤖 Bot (5)                  → /new, /status, /repo, /verbose, /stop
+├── ⚡ superpowers (14)          → brainstorming, TDD, debugging, ...
+├── 📝 commit-commands (3)       → /commit, /commit-push-pr, /clean_gone
+├── 🔍 code-review (2)          → /code-review, /review-pr
+├── 🚀 feature-dev (1)          → executes directly (single item)
+├── 🎨 frontend-design (1)      → executes directly
+├── 📋 claude-md-management (2)  → /revise-claude-md, /claude-md-improver
+├── ⚙️ obsidian-cli (1)          → executes directly (custom skill)
+├── ⚙️ defuddle (1)              → executes directly
+├── ...more custom skills...
+└── 📦 Plugin Store              → Search & install new plugins
+```
+
+Single-item plugins/skills execute directly on tap (no sub-menu).
+
+### Callback Data Convention
+
+```
+Format: "menu:{action}:{argument}"
+
+Navigation:
+  "menu:cat:{plugin_name}"     → show plugin's skills/commands
+  "menu:back"                  → return to top level
+  "menu:back:{plugin_name}"    → return to plugin from detail view
+
+Execution:
+  "menu:run:{item_id}"         → execute a bot command or inject a skill
+
+Plugin management:
+  "menu:plug:{plugin_name}"    → show plugin detail (info, toggle, skills list)
+  "menu:tog:{plugin_name}"     → toggle plugin enable/disable
+  "menu:inst:{plugin_name}"    → install plugin from store
+  "menu:upd:{plugin_name}"     → update plugin to latest version
+
+Store:
+  "menu:store"                 → show plugin store
+  "menu:store:p{page}"         → paginate store results
+```
+
+Note: Telegram limits callback_data to 64 bytes. Plugin/skill names may need truncation + lookup table.
+
+### Data Model
+
+```python
+@dataclass
+class PaletteItem:
+    id: str                    # unique ID, e.g. "superpowers:brainstorming"
+    name: str                  # display name from SKILL.md frontmatter
+    description: str           # from SKILL.md frontmatter
+    action_type: ActionType    # DIRECT_COMMAND | INJECT_SKILL
+    action_value: str          # "/status" or "/commit" or "brainstorming"
+    icon: str                  # emoji (derived or default)
+    source: str                # plugin name or "bot" or "custom"
+    enabled: bool              # cross-referenced with settings.json
+
+class ActionType(Enum):
+    DIRECT_COMMAND = "direct"   # bot handles directly (e.g. /status)
+    INJECT_SKILL = "inject"     # send as text to Claude session
+
+@dataclass
+class PluginInfo:
+    name: str
+    version: str
+    enabled: bool
+    items: list[PaletteItem]   # skills + commands in this plugin
+    path: str                  # filesystem path
+```
+
+### Scanner Implementation
+
+```python
+class CommandPaletteScanner:
+    CLAUDE_DIR = Path.home() / ".claude"
+
+    def scan(self) -> tuple[list[PaletteItem], list[PluginInfo]]:
+        items = []
+        plugins = []
+
+        # 1. Bot commands (hardcoded, always present)
+        items.extend(self._get_bot_commands())
+
+        # 2. Plugin skills: ~/.claude/plugins/cache/claude-plugins-official/*/skills/*/SKILL.md
+        # 3. Plugin commands: ~/.claude/plugins/cache/claude-plugins-official/*/commands/*.md
+        for plugin_dir in self._get_plugin_dirs():
+            plugin_info = self._scan_plugin(plugin_dir)
+            plugins.append(plugin_info)
+            items.extend(plugin_info.items)
+
+        # 4. Custom skills: ~/.claude/skills/*/SKILL.md
+        for skill_dir in self._get_custom_skill_dirs():
+            items.append(self._scan_custom_skill(skill_dir))
+
+        # 5. Cross-reference with settings.json + blocklist.json
+        self._apply_enabled_state(items, plugins)
+
+        return items, plugins
+
+    def _parse_skill_frontmatter(self, path: Path) -> dict:
+        """Parse YAML frontmatter from SKILL.md or command .md file."""
+        # Extract name, description from --- delimited YAML block
+        ...
+```
+
+### Filesystem Paths Scanned
+
+| Path | What | Category |
+|------|------|----------|
+| (hardcoded) | Bot commands: /new, /status, /repo, /verbose, /stop | bot |
+| `~/.claude/plugins/cache/claude-plugins-official/{plugin}/{version}/skills/*/SKILL.md` | Official plugin skills | plugin name |
+| `~/.claude/plugins/cache/claude-plugins-official/{plugin}/{version}/commands/*.md` | Official plugin commands | plugin name |
+| `~/.claude/skills/*/SKILL.md` | Custom skills | custom |
+| `~/.claude/settings.json` | Enabled plugins list | (config) |
+| `~/.claude/plugins/installed_plugins.json` | Plugin versions & metadata | (config) |
+| `~/.claude/plugins/blocklist.json` | Disabled plugins | (config) |
+
+### Action Execution
+
+| Action Type | Behavior |
+|-------------|----------|
+| `DIRECT_COMMAND` | Call the bot's own command handler function directly (e.g., `agentic_status()`) |
+| `INJECT_SKILL` | Send skill name as user text to active Claude session via `ClaudeIntegration.run_command()` |
+
+For skill injection, the bot constructs a message like `/commit` or `/feature-dev` and feeds it to Claude as if the user typed it. This leverages Claude's existing skill invocation mechanism.
+
+### Plugin Management
+
+**Enable/Disable:**
+- Read `~/.claude/plugins/blocklist.json`
+- Add/remove plugin name from blocklist array
+- Write back to file
+- Display confirmation with note: "Takes effect on next /new session"
+
+**Install (Plugin Store):**
+- Query available plugins (TBD: either scrape Claude Code plugin registry or maintain a known-plugins list)
+- Run `claude plugins install {name}` via shell if CLI supports it
+- Otherwise, manual download + write to `installed_plugins.json`
+- Notify user of success/failure
+
+**Update:**
+- Compare installed version (from `installed_plugins.json`) with latest available
+- Run update command or re-download
+
+### Persistent Menu Button
+
+On bot startup, call:
+```python
+await bot.set_chat_menu_button(
+    menu_button=MenuButtonCommands()  # shows / command list
+)
+```
+
+Register `/menu` in `get_bot_commands()` so it appears in Telegram's command autocomplete. The menu button in Telegram opens the command list where `/menu` is the first entry.
+
+Alternative: Use `MenuButtonWebApp` if we ever migrate to a WebApp approach.
+
+## New Files
+
+| File | Purpose |
+|------|---------|
+| `src/bot/features/command_palette.py` | `CommandPaletteScanner`, `PaletteItem`, `PluginInfo`, `ActionType` |
+| `src/bot/handlers/menu.py` | `/menu` command handler, callback handler for menu navigation, plugin management actions |
+
+## Modified Files
+
+| File | Change |
+|------|--------|
+| `src/bot/orchestrator.py` | Register `/menu` handler + menu callback handler in `_register_agentic_handlers()` |
+| `src/bot/orchestrator.py` | Add `/menu` to `get_bot_commands()` |
+| `src/bot/core.py` | Set persistent menu button on startup |
+
+## Callback Data Size Constraint
+
+Telegram limits `callback_data` to 64 bytes. Plugin/skill names can be long (e.g., "claude-md-management:claude-md-improver" = 40 chars + "menu:run:" prefix = 49 chars). Strategy:
+
+- Use short numeric IDs in callback data: `"menu:run:7"`, `"menu:cat:3"`
+- Maintain a per-message mapping dict `{short_id: full_item_id}` stored in `context.user_data`
+- Mapping refreshed on each menu render
+
+## Error Handling
+
+- If `~/.claude/` doesn't exist: show only Bot commands, display note
+- If a SKILL.md has invalid frontmatter: skip it, log warning
+- If plugin toggle fails (permission error): show error inline
+- If skill injection fails (no active session): prompt user to start a session first
+
+## Testing Strategy
+
+- Unit tests for `CommandPaletteScanner` with mock filesystem
+- Unit tests for callback data routing
+- Unit tests for frontmatter parsing (valid, invalid, missing)
+- Integration test for menu navigation flow (mock Telegram API)