← 返回 Skills 市场
388
总下载
1
收藏
1
当前安装
24
版本数
在 OpenClaw 中安装
/install openclaw-sulcus-skill
功能描述
Equip your agent with Sulcus — thermodynamic memory with a knowledge graph. Full SIU pipeline: SIVU (quality gate) → SICU (classifier) → SILU (entity extract...
安全使用建议
This skill appears to be what it says: a memory/knowledge-graph plugin. Before installing: 1) Decide whether you want cloud mode — cloud mode will send memory text, metadata, session events, and embedding requests to a remote Sulcus server using the provided apiKey. 2) Confirm the history import is strictly opt-in and review which files the plugin will read (OPENCLAW_WORKSPACE). 3) If preserving privacy, prefer local-only mode and verify where ~/.sulcus/data is stored/backed up and its permissions. 4) Ask for the plugin's source code or privacy/security documentation and the Sulcus server's data retention and access policies if you plan to use cloud mode. 5) Use a dedicated apiKey with minimal privileges and rotate/revoke it if you stop using the skill.
功能分析
Type: OpenClaw Skill
Name: openclaw-sulcus-skill
Version: 2.5.0
The openclaw-sulcus-skill bundle (specifically SKILL.md) describes an advanced memory system requiring high-privilege capabilities, including outbound network access to api.sulcus.ca and the allowPromptInjection hook for prompt modification. It also requests access to the OPENCLAW_WORKSPACE for history ingestion. While these behaviors are aligned with the stated purpose of a cloud-synced memory system, the combination of broad file access and external data transfer represents a significant attack surface and potential for data exfiltration.
能力标签
能力评估
Purpose & Capability
Name and description match the declared behavior: a memory/graph system that scores, classifies, and injects memories. Declared dependency on an 'openclaw-sulcus' plugin is appropriate for the claimed native/WASM libraries and server interaction; no unrelated binaries or credentials are requested by default.
Instruction Scope
SKILL.md instructs the plugin to persist and auto-inject memories into prompts (before_prompt_build) and describes an opt-in workspace history import. These actions are consistent with a memory-layer plugin but broaden the agent's access (reading workspace history, auto-prepending context to every turn). Confirm that any history import is explicit/opt-in and that automatic injection behavior matches your privacy preferences.
Install Mechanism
Instruction-only skill with no install spec or downloaded artifacts — lowest install risk. The actual binary/native dependencies are expected to be supplied by the referenced plugin, not this SKILL.md.
Credentials
No required environment variables by default; cloud mode optionally uses serverUrl and apiKey which are proportional to sending/storing memories remotely and performing embeddings. The doc notes the same apiKey will be used for embedding requests (BGE-small) and memory storage — this centralization is logical but worth noting because one credential controls multiple flows.
Persistence & Privilege
always:false (no forced global installation). The skill (via plugin) will automatically inject memories into prompts and the platform allows autonomous invocation (disable-model-invocation:false). Automatic injection + cloud mode increases privacy/exfiltration risk if the serverUrl/apiKey are set — consider local-only mode or limiting import/autoinject settings.
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 openclaw-sulcus-skill - 安装完成后，直接呼叫该 Skill 的名称或使用
/openclaw-sulcus-skill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
v2.5.0
v2.5.0: Complete SIRU documentation — adaptive recall weights, recall session logging, feedback API, training pipeline. Updated SIU architecture section with all 5 units (SIVU/SICU/SILU/SIRU/SITU). Context discovery patterns for agents. Cache-friendly injection docs.
v2.4.0
v2.4.0: SIRU (Recall Unit) documentation — adaptive scoring weights, training pipeline, feedback endpoints, all endpoints documented.
v2.3.1
v2.3.1: Published under @digitalforgestudios org.
v2.3.0
v2.3.0: Declare credentials, env vars, and data flows in frontmatter. Add Prerequisites section. Fix config defaults to match plugin v5.7.0. Add importHistory config. Separate local-only and cloud-mode config examples.
v2.2.0
v2.2.0: Multi-signal recall pipeline (semantic+hot+entity+profile), composite scoring, cache-friendly injection (stable confidence bands, no volatile timestamps, deterministic sort, 5-min recall TTL), SIRU training data logging, tokenBudget config.
v3.12.0
Full rewrite for Sulcus v2.2.1
v3.0.0
BREAKING: v3.0.0 — WASM + FFI architecture. Zero spawn, zero network.
v2.0.0
BREAKING: Plugin v2.0.0 removes REST client. MCP-only (zero network). Cloud sync handled by sulcus-sync dylib.
v1.5.3
Version sync 1.5.3. Removed duplicate source files, security annotations.
v1.5.2
Version sync with plugin v1.5.2. autoRecall gating fix reflected in docs.
v1.5.1
Updated slug to openclaw-sulcus-skill. Added links header (ClawHub Skill, ClawHub Plugin, npm, GitHub, Website). Version aligned across all artifacts.
v1.5.0
Fix all scanner flags: version alignment (all files v1.5.0), synced correct manifest (autoRecall/autoCapture default false), removed git-clone install instructions (npm only), removed stale _meta.json. Webhook security warning included.
v1.4.9
Fix default inconsistencies: all examples now show autoRecall/autoCapture as false (matching actual defaults). Added webhook security warning. Zero inconsistencies between docs, examples, and manifest.
v1.4.8
Comprehensive SKILL.md rewrite: reactive triggers, all MCP tools, plugin hooks, auto-recall/auto-capture opt-in, namespace ACL, registration guide, REST API reference, SDK listing. Branding: 'reactive, thermodynamic memory' everywhere.
v1.4.7
Fix scanner flags: autoRecall/autoCapture default to false (opt-in only). Added consentModel, crossNamespaceAccess, consentNote to manifest. Explicit privacy-by-default language throughout.
v1.4.6
Branding: 'Sulcus' everywhere (not 'Sulcus Memory'). Added privacy.dataFlows declaration to plugin manifest. Description emphasizes opt-in, user-configured nature. Addresses ClawHub security scanner concerns.
v1.4.5
Fix: name field matches slug (openclaw-sulcus). Expanded Security & Privacy section with explicit credential requirements, data handling disclosure, tenant isolation details, and disable instructions to address ClawHub security scanner concerns. Bumped to match npm v1.4.5.
v1.4.4
Fix plugin ID: openclaw-sulcus (not memory-sulcus). Updated all references, config examples, and setup docs to match the actual plugin manifest ID. Added plugins.allow guidance. Version aligned with npm @digitalforgestudios/[email protected].
v1.4.0
Remove all cloud sync and webhook references — fully local, zero network. Cloud sync and webhooks are paid features in a separate skill.
v1.3.0
Fix security scan flags: declare binary dependency, clarify local-first architecture, mark cloud sync as optional paid add-on, remove internal IP rules section
元数据
常见问题
Openclaw Sulcus Skill 是什么？
Equip your agent with Sulcus — thermodynamic memory with a knowledge graph. Full SIU pipeline: SIVU (quality gate) → SICU (classifier) → SILU (entity extract... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 388 次。
如何安装 Openclaw Sulcus Skill？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install openclaw-sulcus-skill」即可一键安装，无需额外配置。
Openclaw Sulcus Skill 是免费的吗？
是的，Openclaw Sulcus Skill 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Openclaw Sulcus Skill 支持哪些平台？
Openclaw Sulcus Skill 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 Openclaw Sulcus Skill？
由 Dooley（@mcdoolz）开发并维护，当前版本 v2.5.0。
推荐 Skills