@aemi-cli/aemi 0.0.13 → 0.1.0

package/README.md

@@ -1,5 +1,7 @@
 # AEMI
 
+[한국어](README_KO.md)
+
 Agent Mirror - AI agent routing tool with Telegram and Discord bot integration.
 
 A CLI tool that relays AI agent responses through Telegram/Discord bots.
@@ -90,62 +92,9 @@ Each agent requires its own CLI tool to be installed:
 - **Codex**: `npm install -g @openai/codex`
 - **OpenCode**: `npm install -g opencode` (or see [opencode.ai](https://opencode.ai/docs/) for other methods)
 
-### Integration Feasibility
-
-Each agent CLI provides a non-interactive mode and structured JSON output, making subprocess integration possible.
-
-#### Gemini CLI (Priority 1)
-
-- **Non-interactive mode**: `gemini -p "prompt"` — identical pattern to Claude's `-p` flag
-- **JSON output**: `--output-format json` (single JSON) / `--output-format stream-json` (JSONL stream)
-- **JSON structure**: `{ "response": "...", "stats": {...}, "error": null }`
-- **Stream-json events**: `init`, `message`, `tool_use`, `tool_result`, `error`, `result`
-- **Stability**: Stable release channel (nightly → preview → stable), latest stable v0.29.x
-- **Stdin piping**: Supported (`echo "text" | gemini`)
-- **Auth for subprocess**: `GEMINI_API_KEY` env var (avoids interactive OAuth)
-- **Limitations**: Non-interactive mode restricts tool execution (WriteFile, shell commands require `--yolo/-y`)
-- **Session continuity**: Not supported in non-interactive mode (single-turn only)
-- **Exit codes**: `0` (success), `1` (error), `42` (input error), `53` (turn limit exceeded)
-- **Note**: `-p` flag is deprecated in favor of positional arg (`gemini "prompt"`), but still works
-
-> Gemini CLI has the lowest integration barrier due to its CLI interface being nearly identical to Claude Code.
-> Stream-json event types (`init`, `message`, `tool_use`, `tool_result`) map directly to Claude's `StreamMessage` enum.
-
-#### Codex CLI
-
-- **Non-interactive mode**: `codex exec "prompt"` — uses `exec` subcommand instead of `-p` flag
-- **JSON output**: `codex exec --json "prompt"` → JSONL event stream to stdout
-- **Event types**: `thread.started`, `turn.started`, `turn.completed`, `item.*`, `error`
-- **Auto-approve**: `--full-auto` flag for non-interactive tool execution
-- **Stdin piping**: Supported (`cat prompt.md | codex exec -`)
-- **Session resume**: `codex exec --resume <SESSION_ID> "prompt"`
-- **Extra**: `--output-schema` for schema-constrained responses
-
-> Codex CLI uses a different event model (`item.started`/`item.completed` lifecycle) compared to Claude/Gemini,
-> but maps cleanly to `StreamMessage` via `thread.started`→`Init`, `agent_message`→`Text`, `command_execution`→`ToolUse`/`ToolResult`.
-
-#### OpenCode CLI
-
-- **Non-interactive mode**: `opencode run "prompt"` — uses `run` subcommand
-- **JSON output**: `opencode run --format json "prompt"` → JSONL event stream to stdout
-- **Event types**: `step_start`, `step_finish`, `text`, `tool_use`, `reasoning`, `error`
-- **Tool state tracking**: `tool_use` events include `state.status` (`pending`, `running`, `completed`, `error`)
-- **Session resume**: `opencode run --session <SESSION_ID> "prompt"` / `--continue` for last session
-- **Model selection**: `--model provider/model` for flexible LLM provider switching
-- **File attachment**: `--file <path>` for attaching files to the prompt
-
-> OpenCode uses a part-based event model where `tool_use` events carry state transitions (`pending`→`running`→`completed`).
-> Maps to `StreamMessage` via `text`→`Text`, `tool_use(running)`→`ToolUse`, `tool_use(completed)`→`ToolResult`, `step_finish`→`Done`.
-
-### Implementation Status
-
-- [x] **Extract shared types** — `StreamMessage`, `CancelToken`, `AgentResponse` in `src/services/agent.rs`
-- [x] **Add `src/services/gemini.rs`** — Gemini agent with `-p` and `--output-format stream-json`
-- [x] **Map StreamMessage** — Gemini JSON events → `StreamMessage` enum (`message`→`Text`, `tool_use`→`ToolUse`, `result`→`Done`)
-- [x] **Agent dispatch in bots** — `telegram.rs` and `discord.rs` branch on agent type
-- [x] **Update routing in `main.rs`** — `--agent gemini` accepted alongside `claude`
-- [x] **Add `src/services/codex.rs`** — Codex agent with `exec --json --full-auto`, session resume support
-- [x] **Add `src/services/opencode.rs`** — OpenCode agent with `run --format json`, session resume support
+## Slash Commands
+
+See [docs/slash_commands.md](docs/slash_commands.md) for the full list of bot commands.
 
 ## Supported Platforms
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aemi-cli/aemi",
-  "version": "0.0.13",
+  "version": "0.1.0",
   "description": "LLM CLI routing tool with Telegram and Discord bot integration",
   "license": "MIT",
   "repository": {
@@ -23,9 +23,9 @@
     "aemi": "bin/aemi"
   },
   "optionalDependencies": {
-    "@aemi-cli/linux-x64": "0.0.13",
-    "@aemi-cli/linux-arm64": "0.0.13",
-    "@aemi-cli/darwin-x64": "0.0.13",
-    "@aemi-cli/darwin-arm64": "0.0.13"
+    "@aemi-cli/linux-x64": "0.1.0",
+    "@aemi-cli/linux-arm64": "0.1.0",
+    "@aemi-cli/darwin-x64": "0.1.0",
+    "@aemi-cli/darwin-arm64": "0.1.0"
   }
 }