@sleep2agi/agent-node 2.4.5 → 2.4.6-preview.0

package/README.md

@@ -1,6 +1,6 @@
 # @sleep2agi/agent-node
 
-Agent runtime for Agent Network. Connects to a CommHub server, registers under an alias, and processes incoming tasks with one of three runtimes.
+Agent runtime for Agent Network. Connects to a CommHub server, registers under an alias, and processes incoming tasks with Claude, Codex, Grok Build, or compatible HTTP runtimes.
 
 The supported entry point is the `anet` CLI from `@sleep2agi/agent-network`, which writes the right `config.json`, network token, and environment variables for you.
 
@@ -52,11 +52,39 @@ CLI flags:
 | `claude-agent-sdk` | [@anthropic-ai/claude-agent-sdk](https://www.npmjs.com/package/@anthropic-ai/claude-agent-sdk) | verified | Anthropic-compatible API; works with MiniMax, DeepSeek, GLM, Kimi, Anthropic, OpenRouter, or custom endpoints |
 | `codex-sdk` | [@openai/codex-sdk](https://www.npmjs.com/package/@openai/codex-sdk) | unverified end-to-end | unit tests pass, no full E2E with real codex auth |
 | `claude-code-cli` | local `claude` CLI | unverified end-to-end | runs locally for Claude Pro subscribers (v0.8.2 fixed the session-resume default-loss bug; see [changelog](https://anet.sh/en/changelog)) |
-| `grok-build-acp` | local `grok agent stdio` | engineering preview | requires Grok Build CLI login; stable for receive/reply, session persistence, and explicit CommHub delegation handled by agent-node |
+| `grok-build-acp` | local `grok agent stdio` | stable runtime, native MCP injection boundary remains preview | requires Grok Build CLI login; stable for receive/reply, session persistence, and explicit CommHub delegation handled by agent-node |
 | `http-api` | OpenAI/Anthropic-compatible HTTP | experimental | reads `ANTHROPIC_*`, `OPENAI_*`, or `MINIMAX_CODING_API_KEY` environment variables |
 
 Runtimes are loaded lazily — picking one doesn't pull the others' dependencies. `claude-code-cli` adds zero extra SDK weight.
 
+## Long-running goals for codex-sdk
+
+`codex-sdk` itself does not provide a native `/goal`, timer, cron, or scheduler API. `agent-node` provides an Agent Network scheduler around it:
+
+- send a task that starts with `/goal` or `/loop`
+- the node persists the goal in `.anet/nodes/<name>/goals.json`
+- the running `agent-node` process wakes the runtime on the requested interval
+- each wake asks the agent to check current state, make incremental progress, and report back
+- if the reply contains `目标已完成` or `goal completed`, the goal is marked complete
+
+Example task:
+
+```text
+/goal 每30分钟检查 PR #188 的 CI，修复失败，并在 issue #189 汇报进展
+```
+
+Accepted intervals include `1min`, `5 min`, `hourly`, `daily`, `每5分钟`, `每小时`, and `每天`. Sub-minute intervals such as `30s` or `30秒` are rejected.
+
+When a node starts, its log prints the goal store path:
+
+```text
+goals:   .anet/nodes/<name>/goals.json
+```
+
+Existing running nodes must be restarted after upgrading `agent-node`; Node.js processes do not hot-load the new scheduler code. Goals already written to `goals.json` are reloaded on restart.
+
+Current boundary: there is not yet an `anet goal list/pause/cancel` management CLI or Dashboard goal panel. Use the create reply, node logs, and `goals.json` for now.
+
 ## Grok Build ACP
 
 `grok-build-acp` runs the local Grok Build CLI over Agent Client Protocol:
@@ -70,18 +98,21 @@ anet node start grok-demo
 
 The runtime starts `grok agent stdio`, authenticates with the cached Grok login, opens or loads a Grok session, sends the task prompt, collects streamed ACP notifications, and writes `grokSession` back to the node config.
 
-Preview-stable behavior:
+Stable behavior:
 
 - CommHub task delivery and replies are handled by `agent-node`, not by Grok itself.
 - Plain text tasks should be answered directly by Grok.
-- Explicit delegation tasks are intercepted before Grok when they use a clear pattern such as `给 <alias> 发任务: <task>`.
+- Explicit delegation tasks are intercepted before Grok when they use `给 <alias> 发任务: <task>` or `给 <alias> 派任务: <task>`.
 - Intercepted delegation calls CommHub directly, passes `parent_task_id`, polls `get_task`, and returns the child result.
 
 Known boundary:
 
 - Native Grok MCP tool injection is still experimental. Do not rely on Grok itself seeing `commhub_get_all_status` or `commhub_send_task`.
 - Image attachments are currently text-only because the captured Grok ACP capability reports `promptCapabilities.image=false`.
+- If startup reports `grok CLI not found`, install Grok Build and confirm `grok --version` works.
+- If authentication fails, run `grok` interactively once, then restart the node.
 - `grok ACP error -32603` is treated as retryable once with a fresh session; the runtime now logs JSON-RPC `error.data` when Grok provides it.
+- Grok tool-state boilerplate such as "Do not attempt to use tools from these servers yet" is stripped from final CommHub replies so users see the actual task answer.
 
 ## Provider presets (claude-agent-sdk)