B — Some risk signals Graded October 16, 2026Minor concerns worth a quick check.Based on public adoption and transparency signals as of October 16, 2026. Our opinion, not a code audit or security guarantee.
AI agent skill · Developer tools
khazix-skills
数字生命卡兹克开源的 AI Skills 合集 | Agent Skills: leader（帮你定义目标）, neat-freak 洁癖, hv-analysis, khazix-writer & more — Claude Code, Codex & 40+ agents
Bottom line: Signal-based grade (4/7 points): very widely starred on GitHub, public source code. This reflects adoption and transparency, not a code audit.
- Popularity score
- 48/100
- Rank
- #148 of 500
- GitHub stars
- 21.5k
What it does
khazix-skills is an open-source collection of AI agent skills and capabilities designed for Claude, Codex, and 40+ agent platforms. It includes pre-built skills like leader (goal definition), neat-freak (data cleaning), hv-analysis, and khazix-writer to extend agent functionality without building from scratch.
Who it’s for
AI builders and engineers integrating Claude or multi-agent systems who want production-ready skill modules. Developers comfortable with Chinese documentation and open-source tooling.
Copy-paste setup
Check these steps against the project’s own instructions before running anything.
Verify setup steps against the repository README, as the repo structure is not fully documented here. Likely installation:
git clone https://github.com/KKKKhazix/khazix-skills
cd khazix-skills
npm install
For Claude Desktop MCP integration, check the repo for an mcpServers JSON snippet or configuration guide. Refer to khazix-skills GitHub (opens external site) for exact integration steps.
Risks
- Unclear permission model: No documented scoping of skill capabilities; skills may request broad file system, API, or network access without explicit user approval.
- Prompt injection vulnerability: Skills accepting user input (e.g., leader, writer) may be susceptible to adversarial prompts if input validation is absent.
- Supply-chain risk: Single maintainer (KKKKhazix) with high star count but no visible CI/CD or security audit; updates may not be reviewed by external stakeholders.
- Data exfiltration risk: hv-analysis and other data-processing skills lack transparent logging; sensitive data handling practices are undocumented.
- Maintenance uncertainty: 0 weekly npm downloads and no recent activity signals suggest potential abandonment; critical bugs or security patches may go unaddressed.
How to stay secure
Least-privilege permissions
- Audit each skill's declared permissions before installation; require explicit allowlists for file system paths, API endpoints, and environment variable access.
- Run skills in sandboxed environments with minimal default capabilities; deny all network/file access by default and grant only what's documented as needed.
- Isolate skill execution under separate service accounts with no cross-skill credential sharing; rotate API keys and tokens regularly.
- Review and restrict the
leaderskill's goal-definition scope—it should not be able to modify system policies or invoke unreviewed downstream skills.
Network isolation
- Disable outbound network access for skills unless explicitly required; whitelist only known, trusted API endpoints with certificate pinning where possible.
- Route skill traffic through a monitoring proxy to detect unexpected external connections; block data exfiltration attempts in real time.
- Segregate hv-analysis and data-processing skills onto air-gapped or monitored subnets if handling sensitive information.
Monitoring
- Log all skill inputs, outputs, and permission requests; retain audit trails for forensic review of potential injection or exfiltration incidents.
- Alert on unusual file I/O patterns, large data reads, or repeated failed permission requests that may indicate malicious skill behavior.
- Track npm package updates and maintainer activity; flag use of abandoned or single-maintainer dependencies for immediate review.
Red flags to watch for
- Skills requesting overly broad permissions (e.g.,
fs.*,/etc,process.env) without documented justification—escalate to maintainer before use. - Unexpected network connections or API calls not documented in skill specs; correlate with user input to detect prompt injection exploitation.
- No recent commits, unresponsive maintainer, or zero download metrics—consider forking, auditing, or replacing with actively maintained alternatives.
Alternatives
- Langchain Community Tools – Mature, widely-adopted Python/JS agent skill library with active maintenance and security reviews.
- OpenAI GPT Actions / Custom Tools – Native integration for ChatGPT/API; vendor-controlled permissions and audit logs.
- Anthropic's Claude MCP ecosystem – Official Claude protocol for extending agent capabilities with vetted connectors.
- LlamaIndex Agent Tools – Data-indexing and retrieval focused; well-documented and community-maintained.