seclint

Security and content classifier for AI prompts. Age-appropriate ratings, topic guidance, safety classification. No LLM required.

API

CLI (pipe-friendly)

# Rate: classify a prompt
echo "Help me write a school essay" | seclint rate

# Check: pass/fail threshold
echo "Explain SQL injection attacks" | seclint check --max-rating 12

# Filter: block or sanitize
echo "Some prompt text" | seclint filter --policy strict

HTTP

# Start server
seclint serve 8091

# POST /rate - classify content
curl -X POST http://localhost:8091/rate \
  -H "Content-Type: application/json" \
  -d '{"text":"Help me write a school essay"}'

# POST /check - check threshold
curl -X POST http://localhost:8091/check \
  -H "Content-Type: application/json" \
  -d '{"text":"...", "max_rating":"12"}'

# GET /health - server status
curl http://localhost:8091/health

Output Format

Rate Response

{
  "text": "Help me write a school essay about history",
  "rating": "6+",
  "safe": true,
  "flags": [],
  "confidence": 0.95
}

Check Response

{
  "text": "Explain SQL injection attacks",
  "rating": "16+",
  "max_rating": "12",
  "passes": false,
  "reason": "rating exceeds threshold"
}

Filter Response

{
  "text": "original text...",
  "policy": "strict",
  "action": "block",
  "blocked": true,
  "reason": "explicit content detected"
}

Rating Levels

"6+"      - safe for all ages
"12+"     - mild themes (thriller plots, competitive topics)
"16+"     - mature themes (security, medical, business)
"18+"     - adult only (explicit content, weapons, drugs, violence)
"BLOCKED" - policy violation (illegal activities, harm, exploitation)

Classification Signals

• Keyword matching - domain-specific word lists per rating category
• Context analysis - surrounding words modify severity
• Intent detection - educational vs malicious use
• Combination rules - certain word combinations escalate rating

Integration

As pre-filter in pipeline

# Guard first, then route
echo "prompt" | seclint check --max-rating 16 \
  && echo "prompt" | promptlint analyze \
  | route_to_agent

With ccproxy (Claude Code Proxy)

# ccproxy.yaml
middleware:
  - name: "security_gate"
    type: "seclint"
    endpoint: "http://localhost:8091/check"
    max_rating: "16+"
    action: "block"

With agent pipeline

prompt -> seclint (safe?) -> promptlint (route) -> agent -> archlint (validate) -> costlint (track)
            |
         BLOCK if unsafe

Install

go install github.com/mikeshogin/seclint/cmd/seclint@latest

Policies

• strict - block 16+ and 18+ (safest, most false positives)
• default - block 18+ and explicit violations only
• permissive - only block BLOCKED status (least filtering)

Ecosystem

Part of the AI agent safety and cost optimization ecosystem:

• promptlint - prompt routing by complexity
• costlint - token cost tracking
• archlint - code quality validation

See ECOSYSTEM.md for full integration.

For Humans

Seclint is a content classification tool that rates the maturity level of prompts and other text using age-appropriate categories (6+, 12+, 16+, 18+) without needing to call any LLM. It's designed to be used as a safety gate in AI agent pipelines - you can block or flag prompts before they reach any model, identify content that's beyond a certain maturity level, and integrate it with routing and cost-tracking tools to build a complete safety and optimization system.