Skip to content
Back to skills

Local Sensitive

ASecurity

Routes to local LLM for sensitive operations. Use when user says "로컬에서 처리해줘", "외부로 보내지 마", "민감한 데이터야", "보안 감사해줘", "credential 분석", "시크릿 확인", or handles sensitive data that should not leave the machine.

  • 666 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added February 9, 2026
securitybashapisecurity

Works with

  • api

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill local-sensitive --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Local Sensitive?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Local Sensitive
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-local-sensitive/badge)](https://www.skillsdirectory.com/skills/majiayu000-local-sensitive)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: local-sensitive
description: Routes to local LLM for sensitive operations. Use when user says "로컬에서 처리해줘", "외부로 보내지 마", "민감한 데이터야", "보안 감사해줘", "credential 분석", "시크릿 확인", or handles sensitive data that should not leave the machine.
allowed-tools: Bash(.claude/scripts/local-llm.sh:*), Bash(.claude/scripts/infisical-update.sh:*), Bash(kubectl get:*), Read, Grep, Glob
---

# Local LLM for Sensitive Operations

Route security-sensitive tasks to local LLM to prevent data exfiltration.

## When to Use

- Analyzing secrets or credentials
- Security audit of sensitive code
- Private code review
- Compliance checking with sensitive data
- Any task where data should not leave the machine

## Available Models

| Model | Endpoint | Best For |
|-------|----------|----------|
| `qwen3-coder` | localhost:8003 | Code analysis, generation |
| `gpt-oss` | localhost:8004 | General tasks, tool use |

## How to Use

```bash
.claude/scripts/local-llm.sh qwen3-coder "Review this code for secrets: <code>"
.claude/scripts/local-llm.sh gpt-oss "Analyze security of this config"
```

## Security Guidelines

1. **Never** send sensitive data to external APIs
2. Use this skill when file paths contain: `secret`, `credential`, `.env`, `key`
3. Prefer `qwen3-coder` for code-related tasks
4. Prefer `gpt-oss` for general analysis

## Example Workflow

1. User asks: "Check if there are any hardcoded secrets in this file"
2. Read the file locally
3. Send to local LLM for analysis
4. Report findings without exposing actual secret values

## Infisical Secret Updates

For updating Infisical secrets with sensitive config data (token never exposed):

```bash
# 1. Parse sensitive YAML/JSON with local LLM
.claude/scripts/local-llm.sh qwen3-coder "Parse this config and output JSON: <config>"

# 2. Save to temp file
cat > /tmp/parsed.json << 'EOF'
{"KEY": "value"}
EOF

# 3. Safe update (token fetched internally, never printed)
.claude/scripts/infisical-update.sh safe-parse \
  infisical-app-secrets namespace \
  json /tmp/parsed.json "/path"

# 4. Clean up
rm /tmp/parsed.json
```

See `infisical-manager` skill for detailed workflow.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.