← 返回 Skills 市场
621
总下载
0
收藏
1
当前安装
24
版本数
在 OpenClaw 中安装
/install tide-watch
功能描述
Proactive session capacity monitoring and management for OpenClaw. Prevents context window lockups by warning at configurable thresholds (75%, 85%, 90%, 95%)...
安全使用建议
This skill is coherent with its description: prefer the Directives-Only mode if you want minimal risk (no code installed). If you need the CLI tools: 1) only install the CLI manually (git clone + npm link) if you understand it and trust the repository; 2) confirm the installed version is >= v1.0.1 to avoid the historical CVE; 3) inspect package.json for install hooks (none present) and review lib/resumption.js and lib/capacity.js before linking; 4) verify the CLI only accesses ~/.openclaw/agents/main/sessions/ and not other home files; 5) keep backups and limit auto-backup retention as you prefer. Overall this package appears benign and local-only, but installing any executable CLI should be a conscious, reviewed action.
功能分析
Type: OpenClaw Skill
Name: tide-watch
Version: 1.3.6
The OpenClaw Tide Watch skill is classified as benign. While a critical shell injection vulnerability (CVE-2026-001) was present in version 1.0.0 within `lib/resumption.js`, it was promptly disclosed and fixed in version 1.0.1 by replacing `execSync` with `spawnSync` and `shell: false`, effectively mitigating the RCE risk. The skill explicitly states 'Network Activity: NONE' in `SKILL.md` and `CLAWHUB-SCANNER-RESPONSE.md`, which is confirmed by code review. All file system operations (reading OpenClaw session files, writing resumption prompts, archiving sessions) are local, confined to user-owned OpenClaw directories, and directly aligned with the stated purpose of session capacity monitoring and management. Calls to `child_process.exec` or `execSync` for `openclaw models list` and `openclaw gateway status` are hardcoded and do not incorporate user input, preventing injection. The markdown instructions in `SKILL.md` and other documentation are transparent, guide the AI agent to use its own tools, and do not instruct it to perform malicious actions or exfiltrate data.
能力评估
Purpose & Capability
The skill is described as a hybrid: directives-only monitoring (no code install) plus an optional Node.js CLI for manual management. The manifest, SKILL.md, file list (bin/, lib/), and install spec align with that description. The required config path (~/.openclaw/agents/main/sessions/) is coherent with session-monitoring and archive operations.
Instruction Scope
SKILL.md limits operations to reading and optionally writing OpenClaw session files and resume-prompts in the declared sessions directory. Instructions explicitly separate a safe Directives-Only mode (no code execution) from an optional CLI mode that performs local read/write and archiving. The directives do not instruct the agent to read unrelated sensitive locations or exfiltrate data.
Install Mechanism
Install is an optional local npm install/link of the packaged code (package: "." → npm link). This is an expected and common pattern for CLI tools; it requires manual user action and inspection. No remote arbitrary downloads, no install hooks (package.json scripts only include tests). Risk is moderate only because installing executable code always requires review — the skill documents that and includes source and tests.
Credentials
No credentials requested. The only environment usage is an optional OPENCLAW_SESSION_ID (declared) for auto-detecting the current session; HOME is used via os.homedir() to locate ~/.openclaw. Requested config paths are appropriate for the stated purpose. No unrelated secrets or broad env access are required.
Persistence & Privilege
always:false and disable-model-invocation:false (normal). The skill does not request persistent elevated presence beyond being available as an optional skill; autonomous monitoring only runs if the user adds directives/heartbeats. CLI install is opt-in and does not modify other skills' configs.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install tide-watch - 安装完成后，直接呼叫该 Skill 的名称或使用
/tide-watch触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.3.6
Metadata fix: Declared optional OPENCLAW_SESSION_ID env var. Resolves ClawHub scan finding (BENIGN/SUSPICIOUS → BENIGN/BENIGN expected).
v1.3.5
Documentation fixes: terminology (resume vs restore), comprehensive CLI reference for all flags, updated capabilities.
v1.3.4
Multi-agent aware recommendations + session auto-detection. Recommendations now respect agent boundaries. Added --current flag for heartbeat monitoring efficiency.
v1.3.3
Session-specific archiving: --session flag for archiving individual sessions by ID/label. Supports partial IDs, multiple sessions, and --dry-run preview. Perfect for archiving after saving to memory.
v1.3.2
Human-readable token sizing (18.7k/1M instead of 18,713/1,000,000). Optional --raw-size flag for full precision. Much easier to scan, especially with Gemini models.
v1.3.1
Display name metadata fix: Correct name is 'Tide Watch' (was incorrectly published as 'OpenClaw Tide Watch'). No code changes from v1.3.0.
v1.3.0
Dynamic context limit detection for accurate percentages across all model types (Gemini 1M-2M, Ollama varies). Three-tier fallback: CLI → Config → Defaults. 5-10x accuracy improvement for Gemini models.
v1.2.1
Bug fixes: Archive command multi-agent support (track sessionDir, group by directory). Fixed missing path import. All 113 tests pass. Fixes #31.
v1.2.0
Multi-agent session discovery: Auto-discovers all configured agents from OpenClaw config. Unified dashboard with agent column and per-agent summary. Agent filtering (--agent/--exclude-agent). Backward compatible - zero impact on single-agent users. Fixes #30.
v1.1.6
Hybrid configuration system: CLI flags > env vars > config file > defaults. Configure refresh intervals and gateway timeouts. Secure (0600/0700 permissions), validated (type + range), graceful fallback. Fixes #29.
v1.1.5
Fix gateway status timeout (500ms → 3s). Gateway probe needs time to complete. Async = no blocking. Shows correct status now. Fixes #28.
v1.1.4
Fully async gateway status check. Dashboard refresh always instant (0ms, never blocks). First load shows 'Checking...' then updates. Perfect smooth experience. Fixes #27.
v1.1.3
Fix gateway status blocking dashboard refresh. Cache status for 60s (not every 10s), reduce timeout to 500ms. Dashboard now refreshes instantly. Fixes #26.
v1.1.2
Fix live dashboard UX: eliminate screen flashing + add real-time change tracking. Smooth ANSI cursor positioning, color-coded trend indicators (🔴↑ increasing, 🟢↓ decreasing, 🟡 new). Professional terminal UI. Fixes #25.
v1.1.1
Fix metadata/documentation inconsistency (Fixes #24). Node.js now truly optional - Directives-Only mode requires nothing. Added mode comparison table for clarity. Expected scan: BENIGN.
v1.1.0
Add model display and gateway status monitoring
v1.0.7
Add postinstall script to auto-fix executable permissions after npm install/link
v1.0.6
CRITICAL FIX: Rename bin/tide-watch to bin/tide-watch.js to pass ClawHub text file filter
v1.0.5
Fix ClawHub install spec format to match official spec (kind/label pattern)
v1.0.4
CRITICAL: Fix metadata format for ClawHub registry parser (JSON format instead of nested YAML)
元数据
常见问题
Tide Watch 是什么？
Proactive session capacity monitoring and management for OpenClaw. Prevents context window lockups by warning at configurable thresholds (75%, 85%, 90%, 95%)... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 621 次。
如何安装 Tide Watch？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install tide-watch」即可一键安装，无需额外配置。
Tide Watch 是免费的吗？
是的，Tide Watch 完全免费（开源免费），可自由下载、安装和使用。
Tide Watch 支持哪些平台？
Tide Watch 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Tide Watch？
由 Giddy（@chrisagiddings）开发并维护，当前版本 v1.3.6。
推荐 Skills