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

How to Use

.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):

# 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.