Kivo

KIVO — Agent Knowledge Iteration Engine. A knowledge management system for AI agents that provides knowledge extraction, storage, search, conflict resolution...

Security Scan

The package appears to implement a legitimate agent knowledge engine consistent with its description, but there are implementation and configuration inconsistencies and a few security/operational concerns you should review before installing (weak default admin handling, host-path defaults, and registry metadata mismatch).

Purpose & Capability

The codebase and SKILL.md implement a knowledge engine (SQLite storage, extraction, search, conflict resolution, adapters) consistent with the skill description. However, the registry metadata claimed "instruction-only" / no install spec while the package contains a full library/CLI/web app source tree — this mismatch is noteworthy but plausibly benign (a packaged library with instructions rather than a separate install script). The OpenClawAdapter that defaults storage to ~/.openclaw/… ties the package to a host environment, which is reasonable for a host-embedded mode but should be expected by anyone embedding the skill.

Instruction Scope

SKILL.md instructs normal usage (npm install, init, CLI commands) and lists environment variables. Runtime code will create or modify SQLite DB files, write config (kivo.config.json via npx kivo init), and the bootstrap flow will create an admin user if not present. The bootstrap uses a weak custom simpleHash for the admin account and defaults to password 'admin' if not supplied — that is a functionality + security concern. The OpenClawAdapter will access the user's home directory by default to store the DB. There are no obvious instructions that read arbitrary unrelated host files or contact opaque external endpoints, but file writes to host paths and default weak admin creation are scope-relevant risks.

Install Mechanism

No external download/install spec is embedded in the SKILL manifest; the README/SKILL.md expects users to npm install the package from the npm scope @self-evolving-harness/kivo or build from source. The code imports known packages (better-sqlite3, bcryptjs) and uses no unusual external URL downloads or extract-from-URL installs in the provided sources. This is a normal npm package footprint.

Credentials

SKILL.md documents and the code reference environment variables for DB path, mode, embedding provider and API key, and AUTH_PASSWORD for the web UI — these are reasonable for the stated features. However, the registry metadata lists no required env vars while the code clearly supports/uses many; that mismatch weakens transparency. Also: embedding providers will require external API keys (e.g., openai-style keys); ensure those keys have minimal privilege and are not mistakenly exposed. The bootstrap default admin password and simpleHash are disproportionate and dangerous from a secrets-handling perspective.

Persistence & Privilege

The skill is not always-enabled and is user-invocable; it does not request elevated platform privileges in metadata. At runtime it will create/modify local files (SQLite DB, config) and the OpenClawAdapter defaults to a path under the user's home (~/.openclaw/...); this means installing or running the package will persist data on disk and may interact with host agent state if used in host-embedded mode. That persistence is expected for a knowledge engine but you should be deliberate about DB paths and permissions.

What to consider before installing

This package appears to be a real knowledge engine, but review before running: - Do not run bootstrap with default options on production systems; it will create a default admin account and, if you don't pass adminPassword, uses 'admin'. Provide an adminPassword on initialization and change it immediately. - The bootstrap uses a simpleHash for the initial admin account while other auth code uses bcrypt; check the authentication flow so the created admin will actually authenticate correctly (this is likely an implementation bug you should fix or confirm). - By default the OpenClawAdapter stores the DB under ~/.openclaw/... — if you embed this into a host, explicitly set dbPath/storagePath to a safe, isolated location to avoid overwriting or leaking host agent state. - Embedding/LLM providers require API keys (KIVO_EMBEDDING_API_KEY, etc.). Only provide keys with minimal scope and rotate them if used for testing. Prefer running first in an isolated sandbox or container. - The package bundle is large and contains many modules; if you need to trust it, review the repository (or the included source files) for any network calls or external endpoints beyond documented embedding providers. The manifest/registry metadata mismatch (instruction-only vs full source included) is a transparency issue — ask the publisher for the canonical repository and verify the package integrity before deployment. If you want, I can highlight specific files/functions to inspect (bootstrap admin creation, OpenClawAdapter storage path, and embedding provider implementations) or produce a short remediation checklist you can follow before running.

KIVO — Agent Knowledge Iteration Engine

Agent 知识迭代引擎。为 AI agent 提供知识提取、存储、搜索、冲突解决和迭代学习能力。

Features

• Knowledge extraction and storage (SQLite-backed)
• Semantic and keyword search
• Conflict detection and resolution
• Knowledge distribution and subscription
• Multi-agent authentication and permissions
• Bootstrap initialization and health checks
• Document gate for doc-code consistency

Quick Start

npm install @self-evolving-harness/kivo

import { KnowledgeStore, ExtractionPipeline } from '@self-evolving-harness/kivo';

const store = new KnowledgeStore({ dbPath: './knowledge.db' });
const pipeline = new ExtractionPipeline({ store });
await pipeline.extract(document);

CLI

npx kivo init        # Initialize knowledge base
npx kivo health      # Health check
npx kivo capabilities # Show capabilities