@byteplus/agent-sentry 1.7.0

package/README.md

@@ -0,0 +1,107 @@
+# AgentSentry (OpenClaw Security Plugin)
+
+`agent-sentry` is a security plugin for OpenClaw. It audits LLM traffic and key Agent lifecycle hook points to reduce harmful requests and sensitive data leakage.
+
+## Breaking rename migration notice
+
+This release performs a **breaking external brand rename** from **ClawSentry** to **AgentSentry**. Old external entrypoints are **not compatible** in code after this change.
+
+| Old external entrypoint | New external entrypoint |
+| --- | --- |
+| `claw-sentry` | `agent-sentry` |
+| `plugins.entries.claw-sentry` | `plugins.entries.agent-sentry` |
+| `/clawSentry` | `/agentSentry` |
+| `/cs` | `/as` |
+| `clawSentryListConfig` | `agentSentryListConfig` |
+| `clawSentryListInfo` | `agentSentryListInfo` |
+| `clawSentryListAvailableTools` | `agentSentryListAvailableTools` |
+
+## Requirements
+
+- OpenClaw: `>=2026.4.14`
+- Node.js: `>=22`
+
+## Install
+
+### Volcengine
+
+```bash
+openclaw plugins install @volcengine/agent-sentry
+openclaw config set plugins.entries.agent-sentry.hooks.allowConversationAccess true
+```
+
+### BytePlus
+
+```bash
+openclaw plugins install @byteplus/agent-sentry
+openclaw config set plugins.entries.agent-sentry.hooks.allowConversationAccess true
+```
+
+## Enable
+
+After installing the plugin package, enable it and restart the gateway:
+
+```bash
+openclaw plugins enable agent-sentry
+openclaw gateway restart
+```
+
+## Configuration
+
+Configure the plugin in your OpenClaw config (fields are aligned with `openclaw.plugin.json`):
+
+```yaml
+plugins:
+  entries:
+    agent-sentry:
+      enabled: true
+      config:
+        # Required
+        apiKey: "<encrypted-api-key>"
+        clawId: "<claw-id>"
+        ctEndpoint: "https://<control-endpoint>"
+        lmEndpoint: "https://<moderation-endpoint>"
+
+        # Optional
+        asClawId: "<as-claw-id>"
+        configVersion: "<version-string>"
+        openClawDir: "<path-to-openclaw-state-dir>"
+
+        # Behavior
+        logRecord: false
+        # enableFetch is deprecated since 1.5.0-dlp and is now a no-op; DLP runs
+        # via before_dispatch + before_prompt_build (see dlpInputGate below).
+        enableFetch: false
+        enableBeforeToolCall: true
+        enableAfterToolCall: true
+        enableHeartbeat: true
+        timeoutMs: 30000
+        failureThreshold: 3
+        retryInterval: 60
+        maxRetryInterval: 3600
+        # Optional input-side DLP enforcement mode for before_dispatch.
+        # 'marker_only' (default) keeps the user message and lets
+        # before_prompt_build inject a system reminder. 'direct_block' returns
+        # {handled:true, text:securityReason} from before_dispatch.
+        dlpInputGate:
+          mode: marker_only
+        language: "en" # or "zh"
+```
+
+## Hooks & auditing
+
+- `before_dispatch` + `before_prompt_build`: collect moderation results and inject safe context into prompts.
+- `before_tool_call`: audit tool name/params (and optional tool-result context). The `isGroup` flag is derived from `ctx.sessionKey` (format `agent:<id>:<channel>:group|direct:<target>`).
+- `after_tool_call`: store tool result context for subsequent audits.
+- `llm_input`: capture LLM input (e.g. system prompt) for auditing/traceability.
+- `agent_end`: upload conversation record and clear per-session cached contexts.
+- Degradation/circuit breaker: automatically bypasses checks when the security service is unavailable and probes for recovery.
+
+## Development
+
+```bash
+pnpm -C openclaw-plugin build
+pnpm -C openclaw-plugin test
+```
+
+Publishing scripts are defined in `openclaw-plugin/package.json` (including `beta` and `private` tags).

package/RELEASE_NOTES.md

@@ -0,0 +1,24 @@
+# AgentSentry rename release notes
+
+This release introduces a **breaking external rename** from **ClawSentry** to **AgentSentry**.
+
+Old external entrypoints are **no longer compatible** after upgrading.
+
+| Old external entrypoint | New external entrypoint |
+| --- | --- |
+| `claw-sentry` | `agent-sentry` |
+| `plugins.entries.claw-sentry` | `plugins.entries.agent-sentry` |
+| `/clawSentry` | `/agentSentry` |
+| `/cs` | `/as` |
+| `clawSentryListConfig` | `agentSentryListConfig` |
+| `clawSentryListInfo` | `agentSentryListInfo` |
+| `clawSentryListAvailableTools` | `agentSentryListAvailableTools` |
+
+## Required migration steps
+
+1. Reinstall or reference the plugin package as `agent-sentry`.
+2. Update OpenClaw config keys from `plugins.entries.claw-sentry` to `plugins.entries.agent-sentry`.
+3. Update slash command usage from `/clawSentry` and `/cs` to `/agentSentry` and `/as`.
+4. Update any tool integrations to the `agentSentry*` identifiers.
+
+Protocol compatibility fields such as `clawId`, `ClawID`, `claw_id`, `X-Ai-Claw-Id`, and `pluginType=clawsentry` remain unchanged.

package/agent-sentry/SKILL.md

@@ -0,0 +1,53 @@
+---
+name: agent-sentry
+description: |
+  AI Assistant Security (AgentSentry) 安全防护系统，用于查看防护配置、查看配置信息、查看可用工具等。
+  Activate when user needs to: 查看防护配置、查看配置信息、查看可用工具、查看AgentSentry可用工具、查看ai-assistant-security-openclaw、查看安全防护、查看安全配置、使用安全工具、打开安全工具、调用安全工具、安全防护工具、安全工具、使用防护工具、打开防护工具、调用防护工具、防护工具。
+metadata:
+  {
+    "openclaw":
+      {
+        "emoji": "🛡️",
+        "skillKey": "agent-sentry",
+        "requires": { "config": ["plugins.entries.agent-sentry.enabled"] },
+      },
+  }
+---
+
+# AgentSentry
+
+AgentSentry 是一个安全防护系统，提供 LLM 和 Agent 的安全防护功能。
+
+## Migration notice
+
+这是一次破坏性外显改名，旧入口不再兼容：
+
+| 旧入口 | 新入口 |
+| --- | --- |
+| `claw-sentry` | `agent-sentry` |
+| `plugins.entries.claw-sentry` | `plugins.entries.agent-sentry` |
+| `/clawSentry` | `/agentSentry` |
+| `/cs` | `/as` |
+| `clawSentryListConfig` | `agentSentryListConfig` |
+| `clawSentryListInfo` | `agentSentryListInfo` |
+| `clawSentryListAvailableTools` | `agentSentryListAvailableTools` |
+
+## Agent rules:
+
+- Call tools directly. Never suggest slash commands to the user.
+- Always inform the user if they don't have permission to use a tool.
+- When a tool returns "no permission", tell the user clearly and remind them they can ask "我可以使用什么功能" or "查看可用工具" to see what tools are available for their version.
+
+## Tool Parameters
+
+### agentSentryListConfig
+
+查看防护配置。**Call when:** "查看防护配置", "看一下防护规则", "有哪些防护规则", "查看安全配置", "查看安全防护配置", "看一下当前配置". No params.
+
+### agentSentryListInfo
+
+查看配置信息。**Call when:** "查看配置信息", "看一下配置", "有哪些配置", "查看插件配置", "看一下插件配置", "查看AgentSentry配置", "看一下AgentSentry配置", "查看配置", "查看版本信息", "看一下版本信息". No params.
+
+### agentSentryListAvailableTools
+
+查看可用工具。**Call when:** "查看可用工具", "我能用哪些工具", "查看AgentSentry可用工具", "查看当前版本可以使用的所有工具列表", "有哪些工具可以用", "查看可用命令", "我可以用什么工具", "有什么功能", "我能做什么", "有哪些功能", "查看功能", "我可以使用哪些功能", "有什么工具", "查看我能使用的工具", "当前版本有什么功能", "有什么可用的功能", "我能使用什么", "我有什么功能可以用". No params.