hajimi-claw 0.1.0 → 0.1.2

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Dongshan_Randeng
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,30 +1,36 @@
 # hajimi-claw
 
-Single-user Telegram-first ops agent in Rust.
+Single-user Telegram/Feishu-first ops agent in Rust.
 
 ## Current scope
 
 - Telegram channel -> gateway -> runtime command flow
+- Feishu webhook channel -> gateway -> runtime command flow
 - Telegram long polling command surface
 - Single active task gate
 - Structured tools for file access, Docker, and systemd
-- Guarded local command execution with approvals and short-lived elevated lease
+- Guarded local command execution with approval mode, short-lived elevated mode, and full elevated mode
 - SQLite audit/task/session persistence
 - Windows-safe execution mode with allowlist checks and Job Object cleanup
-- Telegram onboarding flow for custom providers and chat-level provider binding
+- Channel-aware onboarding for Telegram or Feishu plus provider/model setup
+- Configurable multi-agent orchestration with coordinator/worker/integrator flow
 
 ## Running
 
-1. Copy `config.example.toml` to `config.toml`.
-2. Fill in the Telegram bot token and admin ids.
-3. Set the provider secret key env var before starting:
-   `set HAJIMI_CLAW_MASTER_KEY=replace-me-with-a-long-random-string`
-4. Optional: fill in the `llm` section to bootstrap a default provider on first start.
-5. Run `cargo run`.
-6. In Telegram, use `/onboard` to add or switch providers interactively.
+1. Run `cargo run -- onboard` or `hajimi onboard`.
+2. `hajimi onboard` lets you choose `telegram`, `feishu`, or `skip`, verifies the channel credentials when configured, and then interactively guides provider model selection.
+3. Start the daemon with `cargo run` or `hajimi`.
+4. In Telegram, use `/menu` for the main quick-action panel, or just send plain text and let hajimi reply in natural language.
+5. In Feishu mode, configure the app's event subscription URL to point at `http(s)://<your-host><event_path>` for the local webhook listener.
 
 Set `HAJIMI_CLAW_CONFIG` if you want to load a different config path.
 
+For background use without keeping a terminal open:
+
+- `hajimi launch`
+- `hajimi status`
+- `hajimi stop`
+
 ## Install
 
 ### Windows
@@ -34,7 +40,7 @@ Set `HAJIMI_CLAW_CONFIG` if you want to load a different config path.
 - System-wide install:
   `powershell -ExecutionPolicy Bypass -File .\scripts\install.ps1 -System`
 
-The installer copies `hajimi-claw.exe` and `config.example.toml`, and updates `PATH`.
+The installer copies `hajimi-claw.exe`, creates the `hajimi.exe` alias, copies `config.example.toml`, and updates `PATH`.
 
 ### Linux
 
@@ -45,7 +51,7 @@ The installer copies `hajimi-claw.exe` and `config.example.toml`, and updates `P
 - System-wide install with systemd unit:
   `sudo sh ./scripts/install.sh --system --install-service`
 
-The Linux installer copies the binary to `PREFIX/bin` and `config.example.toml` to `PREFIX/share/hajimi-claw`.
+The Linux installer copies the binary to `PREFIX/bin/hajimi-claw`, creates the `PREFIX/bin/hajimi` alias, and copies `config.example.toml` to `PREFIX/share/hajimi-claw`.
 
 ### npm
 
@@ -54,20 +60,203 @@ The Linux installer copies the binary to `PREFIX/bin` and `config.example.toml`
 - Global install from the repo:
   `npm install -g .`
 
-The npm package builds the Rust binary locally during `postinstall`, so `cargo` must already be installed on the target machine.
+The npm package exposes both `hajimi` and `hajimi-claw`, but it no longer compiles Rust during
+install. Instead:
+
+- The root package is a small launcher package
+- Platform-specific binaries are published as optional npm packages
+- `npm install -g hajimi-claw` installs the matching prebuilt binary package when one is available
+
+Supported prebuilt package names:
+
+- `hajimi-claw-win32-x64-msvc`
+- `hajimi-claw-win32-arm64-msvc`
+- `hajimi-claw-linux-x64-gnu`
+- `hajimi-claw-linux-arm64-gnu`
+
+For local testing from the repo, stage a binary into the local platform package first:
+
+```bash
+cargo build --release
+npm run stage:binary
+npm install -g .
+```
+
+For CI or release staging, you can target a specific package explicitly:
+
+```bash
+node scripts/stage-prebuilt.js win32-x64-msvc path/to/hajimi-claw.exe
+node scripts/stage-prebuilt.js linux-x64-gnu path/to/hajimi-claw
+```
+
+Recommended publish order:
+
+1. Build each platform binary in CI
+2. Stage it into the matching `npm/platforms/<target>/bin/`
+3. Publish each platform package
+4. Publish the root `hajimi-claw` package last
+
+Current GitHub Actions automation is wired for:
+
+- `hajimi-claw-win32-x64-msvc`
+- `hajimi-claw-linux-x64-gnu`
+
+The arm64 package skeletons are present in the repo, but they are not yet part of the automated
+publish matrix.
+
+## npm Versioning
+
+For npm releases, the root `package.json` is the single version source.
+
+When you change `package.json.version`, run:
+
+```bash
+npm run sync:npm-versions
+```
+
+That script updates:
+
+- the root package `optionalDependencies`
+- every platform package version under `npm/platforms/*/package.json`
+
+The release workflow also runs this sync step automatically, so you do not need to edit platform
+package versions by hand.
+
+## Channels
+
+- `telegram`
+- `feishu`
+
+`hajimi onboard` now lets you pick the primary channel:
+
+- `telegram`: verifies the bot token and can auto-pair the admin user/chat
+- `feishu`: asks for `app_id` and `app_secret`, verifies them by requesting a tenant access token, and supports both `webhook` and `long-connection` modes
+- `skip`: leaves channel setup empty for now so you can finish provider/model setup first
+
+Current Feishu limitations:
+
+- In `webhook` mode, you still need to configure the Feishu event subscription URL manually in the Feishu developer console
+- In `long-connection` mode, message receive events can come through the WebSocket connection, but card button callbacks still need the lightweight HTTP callback path
+- Feishu card support currently targets `menu/provider/model` style button flows; it is not a full parity implementation of Telegram inline interactions
+
+## CLI
+
+- `hajimi`
+- `hajimi daemon`
+- `hajimi launch`
+- `hajimi stop`
+- `hajimi status`
+- `hajimi onboard`
+- `hajimi providers`
+- `hajimi provider current`
+- `hajimi provider use <provider-id>`
+- `hajimi provider models [provider-id]`
+- `hajimi provider set-model <provider-id> <model>`
+- `hajimi model current`
+- `hajimi model use [model]`
+- `hajimi models [provider-id]`
+- `hajimi restart`
+- `hajimi help`
+
+## Multi-Agent
+
+`hajimi` can split one natural-language request into multiple sub-agents. This is configured in
+`config.toml` under `[multi_agent]`:
+
+```toml
+[multi_agent]
+enabled = true
+auto_delegate = false
+default_workers = 3
+max_workers = 8
+worker_timeout_secs = 90
+max_context_chars_per_worker = 24000
+```
+
+Natural-language triggers:
+
+- `use 4 agents to analyze this`
+- `开 6 个 agent 帮我排查`
+- `please use sub agents for this`
+
+Behavior:
+
+- If the prompt explicitly asks for `N agents` or `N workers`, hajimi uses that count up to
+  `max_workers`
+- If the prompt says `sub agent` / `multi agent`, hajimi uses `default_workers`
+- `auto_delegate = true` enables light automatic delegation for obviously parallel analysis tasks
+- Worker agents only perform parallel reasoning in v1; local command execution still stays behind
+  the existing single runtime/policy path
+- Session-level controls are available in chat:
+  - `/agents on`
+  - `/agents off`
+  - `/agents auto`
+  - `/agents status`
+- When a natural-language task is about to run in multi-agent mode, Telegram now shows a temporary
+  progress card and Feishu sends a status card before the final reply
+
+## Persona Files
+
+`hajimi onboard` now creates empty persona files in `~/.hajimi/`:
+
+- `soul.md`
+- `agents.md`
+- `tools.md`
+- `skills.md`
+
+`hajimi` reloads these files on each request, so Telegram edits take effect on the next `/ask`.
+
+- Auto-discovered files: `soul.md`, `agents.md`, `AGENTS.md`, `tools.md`, `skills.md`
+- Search roots: the current working directory, the config directory, and `~/.hajimi`
+- Optional explicit list: set `[persona].prompt_files` in `config.toml`
+
+Use these files for:
+
+- `soul.md`: tone, temperament, and high-level persona
+- `agents.md` or `AGENTS.md`: repo or operator instructions
+- `tools.md`: tool-use policy and operational preferences
+- `skills.md`: extra habits, playbooks, and skill-selection hints
 
 ## Telegram commands
 
 - `/onboard`
 - `/onboard cancel`
 - `/provider list`
+- `/provider add`
 - `/provider current`
 - `/provider use <id>`
 - `/provider bind <id>`
 - `/provider test [id]`
 - `/provider models [id]`
+- `/provider set-model <provider-id> <model>`
+- `/model current`
+- `/model use [model]`
+- `/persona list`
+- `/persona read <soul|agents|tools|skills>`
+- `/persona write <file> <content>`
+- `/persona append <file> <content>`
 - `/ask <text>`
 - `/shell open [name]`
 - `/shell exec <cmd>`
 - `/shell close`
 - `/status`
+- `/menu`
+- `/elevated on`
+- `/elevated off`
+- `/elevated ask`
+- `/elevated full`
+
+Plain Telegram text now defaults to a natural-language task, so `/ask` is optional for normal requests. hajimi sends a Telegram `typing` action, posts a short placeholder, and then edits that message with the final answer.
+
+The bot also registers Telegram slash commands with `setMyCommands` on startup and exposes an inline quick-action menu via `/menu` and `/help`. `/provider current` and `/model current` now return inline buttons so you can switch provider/model directly from Telegram without typing full commands.
+
+## Provider And Model Switching
+
+- Add a new provider: `hajimi onboard` or Telegram `/provider add`
+- List configured providers: `hajimi providers` or Telegram `/provider list`
+- Switch the default provider: `hajimi provider use <provider-id>` or Telegram `/provider use <id>`
+- See the current model: `hajimi model current` or Telegram `/model current`
+- Switch the current model on the active provider: `hajimi model use [model]` or Telegram `/model use [model]`
+- Open the model picker with inline buttons: `hajimi model use` or Telegram `/model use`
+- Switch a specific provider to a specific model: `hajimi provider set-model <provider-id> <model>` or Telegram `/provider set-model <provider-id> <model>`
+- `hajimi onboard` now supports optional fallback models per provider; if the primary model fails, hajimi will retry the provider using the configured fallback models in order

package/bin/hajimi-claw.js

@@ -6,11 +6,85 @@ const { spawn } = require("child_process");
 
 const root = path.resolve(__dirname, "..");
 const binaryName = process.platform === "win32" ? "hajimi-claw.exe" : "hajimi-claw";
-const binaryPath = path.join(root, "npm-dist", binaryName);
 
-if (!fs.existsSync(binaryPath)) {
+function detectLinuxLibc() {
+  if (process.platform !== "linux") {
+    return null;
+  }
+  const report = process.report && typeof process.report.getReport === "function"
+    ? process.report.getReport()
+    : null;
+  const glibc = report && report.header && report.header.glibcVersionRuntime;
+  return glibc ? "gnu" : "musl";
+}
+
+function platformPackageName() {
+  if (process.platform === "win32" && process.arch === "x64") {
+    return "hajimi-claw-win32-x64-msvc";
+  }
+  if (process.platform === "win32" && process.arch === "arm64") {
+    return "hajimi-claw-win32-arm64-msvc";
+  }
+  const libc = detectLinuxLibc();
+  if (process.platform === "linux" && process.arch === "x64" && libc === "gnu") {
+    return "hajimi-claw-linux-x64-gnu";
+  }
+  if (process.platform === "linux" && process.arch === "arm64" && libc === "gnu") {
+    return "hajimi-claw-linux-arm64-gnu";
+  }
+  return null;
+}
+
+function resolveInstalledBinary() {
+  const packageName = platformPackageName();
+  if (!packageName) {
+    return null;
+  }
+  try {
+    const entry = require.resolve(packageName, { paths: [root] });
+    const binaryPath = require(entry);
+    if (typeof binaryPath === "string" && fs.existsSync(binaryPath)) {
+      return binaryPath;
+    }
+  } catch (_) {
+    return null;
+  }
+  return null;
+}
+
+function resolveLocalFallback() {
+  const localDirMap = {
+    "hajimi-claw-win32-x64-msvc": "win32-x64-msvc",
+    "hajimi-claw-win32-arm64-msvc": "win32-arm64-msvc",
+    "hajimi-claw-linux-x64-gnu": "linux-x64-gnu",
+    "hajimi-claw-linux-arm64-gnu": "linux-arm64-gnu",
+  };
+  const packageName = platformPackageName();
+  const candidates = [];
+  if (process.env.HAJIMI_CLAW_BINARY_PATH) {
+    candidates.push(process.env.HAJIMI_CLAW_BINARY_PATH);
+  }
+  if (packageName && localDirMap[packageName]) {
+    candidates.push(
+      path.join(root, "npm", "platforms", localDirMap[packageName], "bin", binaryName)
+    );
+  }
+  candidates.push(path.join(root, "npm-dist", binaryName));
+  candidates.push(path.join(root, "target", "release", binaryName));
+  return candidates.find((candidate) => fs.existsSync(candidate)) || null;
+}
+
+const binaryPath = resolveInstalledBinary() || resolveLocalFallback();
+
+if (!binaryPath) {
+  const packageName = platformPackageName();
+  const targetLabel = packageName || `${process.platform}-${process.arch}`;
   console.error(
-    "hajimi-claw binary is not installed. Reinstall the npm package or run `npm rebuild hajimi-claw`."
+    [
+      `No prebuilt hajimi binary is available for ${targetLabel}.`,
+      "Install a published platform package, or stage a local binary with `npm run stage:binary`.",
+      "If you are publishing releases, publish the matching platform package before the root package."
+    ].join("\n")
   );
   process.exit(1);
 }

package/config.example.toml

@@ -1,7 +1,20 @@
+[channel]
+kind = "telegram"
+
 [telegram]
 bot_token = "123456:replace-me"
 poll_timeout_secs = 30
 
+[feishu]
+# `mode = "webhook"`: expose listen_addr + event_path through your reverse proxy or public host.
+# `mode = "long-connection"`: message events come from Feishu WebSocket; keep card_callback_path reachable if you want card button callbacks.
+app_id = ""
+app_secret = ""
+mode = "webhook"
+listen_addr = "0.0.0.0:8787"
+event_path = "/feishu/events"
+card_callback_path = "/feishu/card"
+
 [llm]
 # Optional bootstrap provider written into SQLite on first start.
 base_url = "https://api.openai.com/v1"
@@ -9,6 +22,9 @@ api_key = "replace-me"
 model = "gpt-4.1-mini"
 static_fallback_response = "LLM backend not configured."
 
+# Provider fallback models are stored in SQLite per provider by `hajimi onboard`.
+# Example: gpt-4o-mini -> gpt-4.1-mini -> gpt-4.1
+
 [storage]
 sqlite_path = "./data/hajimi-claw.sqlite3"
 
@@ -19,11 +35,26 @@ master_key_env = "HAJIMI_CLAW_MASTER_KEY"
 [execution]
 mode = "auto"
 
+[multi_agent]
+enabled = true
+auto_delegate = false
+default_workers = 3
+max_workers = 8
+worker_timeout_secs = 90
+max_context_chars_per_worker = 24000
+
+[persona]
+# hajimi onboard creates soul.md, agents.md, tools.md, and skills.md here.
+directory = "~/.hajimi"
+# Optional. If empty, hajimi auto-loads soul.md, agents.md, AGENTS.md, tools.md,
+# and skills.md from the current working directory, the config directory, and persona.directory.
+prompt_files = []
+
 [policy]
 admin_user_id = 123456789
 admin_chat_id = 123456789
-allowed_workdirs = ["./", "./data", "C:/temp"]
-writable_workdirs = ["./data", "C:/temp"]
+allowed_workdirs = ["./", "./data", "~/.hajimi", "C:/temp"]
+writable_workdirs = ["./data", "~/.hajimi", "C:/temp"]
 windows_safe_allowlist = ["cmd", "cmd.exe", "powershell", "powershell.exe", "pwsh", "pwsh.exe", "docker", "docker.exe", "systemctl"]
 guarded_patterns = ["\\b(systemctl|docker)\\s+(restart|stop|rm)\\b", "\\b(chmod|chown|mv|cp)\\b"]
 dangerous_patterns = ["\\b(rm|del)\\s+(-rf|/s|/q|/f)\\b", "\\b(sudo|su|passwd|shutdown|reboot)\\b", ">\\s*/", "format\\s+"]

package/npm/platforms/linux-arm64-gnu/LICENSE

@@ -0,0 +1 @@
+MIT

package/npm/platforms/linux-arm64-gnu/README.md

@@ -0,0 +1,3 @@
+# hajimi-claw-linux-arm64-gnu
+
+Prebuilt Linux arm64 glibc binary package for `hajimi-claw`.

package/npm/platforms/linux-arm64-gnu/bin/.gitkeep

@@ -0,0 +1 @@
+

package/npm/platforms/linux-arm64-gnu/index.js

@@ -0,0 +1,3 @@
+const path = require("path");
+
+module.exports = path.join(__dirname, "bin", "hajimi-claw");

package/npm/platforms/linux-arm64-gnu/package.json

@@ -0,0 +1,21 @@
+{
+  "name": "hajimi-claw-linux-arm64-gnu",
+  "version": "0.1.2",
+  "description": "Prebuilt Linux arm64 glibc binary for hajimi-claw",
+  "os": [
+    "linux"
+  ],
+  "cpu": [
+    "arm64"
+  ],
+  "libc": [
+    "glibc"
+  ],
+  "files": [
+    "bin",
+    "index.js",
+    "README.md",
+    "LICENSE"
+  ],
+  "license": "MIT"
+}

package/npm/platforms/linux-x64-gnu/LICENSE

@@ -0,0 +1 @@
+MIT

package/npm/platforms/linux-x64-gnu/README.md

@@ -0,0 +1,3 @@
+# hajimi-claw-linux-x64-gnu
+
+Prebuilt Linux x64 glibc binary package for `hajimi-claw`.

package/npm/platforms/linux-x64-gnu/bin/.gitkeep

@@ -0,0 +1 @@
+

package/npm/platforms/linux-x64-gnu/index.js

@@ -0,0 +1,3 @@
+const path = require("path");
+
+module.exports = path.join(__dirname, "bin", "hajimi-claw");

package/npm/platforms/linux-x64-gnu/package.json

@@ -0,0 +1,21 @@
+{
+  "name": "hajimi-claw-linux-x64-gnu",
+  "version": "0.1.2",
+  "description": "Prebuilt Linux x64 glibc binary for hajimi-claw",
+  "os": [
+    "linux"
+  ],
+  "cpu": [
+    "x64"
+  ],
+  "libc": [
+    "glibc"
+  ],
+  "files": [
+    "bin",
+    "index.js",
+    "README.md",
+    "LICENSE"
+  ],
+  "license": "MIT"
+}

package/npm/platforms/win32-arm64-msvc/LICENSE

@@ -0,0 +1 @@
+MIT

package/npm/platforms/win32-arm64-msvc/README.md

@@ -0,0 +1,3 @@
+# hajimi-claw-win32-arm64-msvc
+
+Prebuilt Windows arm64 binary package for `hajimi-claw`.

package/npm/platforms/win32-arm64-msvc/bin/.gitkeep

@@ -0,0 +1 @@
+

package/npm/platforms/win32-arm64-msvc/index.js

@@ -0,0 +1,3 @@
+const path = require("path");
+
+module.exports = path.join(__dirname, "bin", "hajimi-claw.exe");

package/npm/platforms/win32-arm64-msvc/package.json

@@ -0,0 +1,18 @@
+{
+  "name": "hajimi-claw-win32-arm64-msvc",
+  "version": "0.1.2",
+  "description": "Prebuilt Windows arm64 binary for hajimi-claw",
+  "os": [
+    "win32"
+  ],
+  "cpu": [
+    "arm64"
+  ],
+  "files": [
+    "bin",
+    "index.js",
+    "README.md",
+    "LICENSE"
+  ],
+  "license": "MIT"
+}

package/npm/platforms/win32-x64-msvc/LICENSE

@@ -0,0 +1 @@
+MIT

package/npm/platforms/win32-x64-msvc/README.md

@@ -0,0 +1,3 @@
+# hajimi-claw-win32-x64-msvc
+
+Prebuilt Windows x64 binary package for `hajimi-claw`.

package/npm/platforms/win32-x64-msvc/bin/.gitkeep

@@ -0,0 +1 @@
+

package/npm/platforms/win32-x64-msvc/index.js

@@ -0,0 +1,3 @@
+const path = require("path");
+
+module.exports = path.join(__dirname, "bin", "hajimi-claw.exe");

package/npm/platforms/win32-x64-msvc/package.json

@@ -0,0 +1,18 @@
+{
+  "name": "hajimi-claw-win32-x64-msvc",
+  "version": "0.1.2",
+  "description": "Prebuilt Windows x64 binary for hajimi-claw",
+  "os": [
+    "win32"
+  ],
+  "cpu": [
+    "x64"
+  ],
+  "files": [
+    "bin",
+    "index.js",
+    "README.md",
+    "LICENSE"
+  ],
+  "license": "MIT"
+}

package/package.json

@@ -1,26 +1,33 @@
 {
   "name": "hajimi-claw",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "Single-user Telegram-first ops agent in Rust",
   "bin": {
+    "hajimi": "bin/hajimi-claw.js",
     "hajimi-claw": "bin/hajimi-claw.js"
   },
+  "preferGlobal": true,
+  "license": "MIT",
+  "engines": {
+    "node": ">=18"
+  },
+  "optionalDependencies": {
+    "hajimi-claw-linux-arm64-gnu": "0.1.2",
+    "hajimi-claw-linux-x64-gnu": "0.1.2",
+    "hajimi-claw-win32-arm64-msvc": "0.1.2",
+    "hajimi-claw-win32-x64-msvc": "0.1.2"
+  },
   "files": [
-    "Cargo.toml",
-    "Cargo.lock",
-    "src",
-    "crates",
     "bin",
-    "scripts/npm-install.js",
+    "npm/platforms",
+    "scripts/sync-npm-versions.js",
+    "scripts/stage-prebuilt.js",
     "config.example.toml",
-    "README.md"
+    "README.md",
+    "LICENSE"
   ],
   "scripts": {
-    "postinstall": "node scripts/npm-install.js"
-  },
-  "preferGlobal": true,
-  "license": "MIT",
-  "engines": {
-    "node": ">=18"
+    "sync:npm-versions": "node scripts/sync-npm-versions.js",
+    "stage:binary": "npm run sync:npm-versions && node scripts/stage-prebuilt.js"
   }
 }