CostGuardAI CLI

Preflight analysis for LLM prompts. Catch cost overruns and failure risk before they ship.

npm install -g @camj78/costguardai
costguardai ci --fail-on-risk 70

Install

npm install -g @camj78/costguardai

Usage

# Analyze a prompt
costguardai analyze my-prompt.txt

# Block unsafe prompts in CI
costguardai ci --fail-on-risk 70

# Initialize CostGuardAI in your repo
costguardai init

CostGuardAI CI will fail if the prompt's CostGuardAI Safety Score falls below your configured threshold.

Automatic Prompt Hardening

costguardai fix prompt.txt

CostGuardAI can automatically harden prompts by adding injection protections, system boundaries, and safety guards.

CostGuardAI Fix Mode
────────────────────────────────────────────────
Original Safety Score: 58
Issues Detected:
  • Structural Risk
  • Ambiguity Risk

Applying prompt hardening...
  ✔ Added system boundary protections
  ✔ Removed injection vulnerability
  ✔ Added explicit instruction scoping

New Safety Score: 92

Hardened prompt written to:
  prompt.hardened.txt

The hardened prompt is written to <original-name>.hardened.<ext> alongside the original file.

GitHub Actions

name: CostGuardAI
on: [pull_request]
jobs:
  scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: npm install -g @camj78/costguardai
      - run: costguardai ci --fail-on-risk 70

Exit Codes

| Code | Meaning | |------|---------| | 0 | All files below threshold | | 1 | Runtime error | | 2 | One or more files at or above threshold |

Commands

| Command | Description | |---------|-------------| | analyze <path> | Analyze prompt files in a directory or single file | | fix <file> | Harden a prompt and improve its safety score | | ci [path] | CI-native scan with --fail-on-risk <n> | | init | Create costguard.config.json with defaults |

Options

--model <id>             gpt-4o-mini (default), gpt-4o, claude-sonnet-4-6, claude-haiku-4-5
--format text|md|json    Output format (default: text)
--json                   Shorthand for --format json
--fail-on-risk <n>       Exit 2 if any file risk_score >= n  [ci command]
--threshold <n>          Same as --fail-on-risk               [analyze command]
--ext <exts>             Comma-separated extensions (default: .txt,.md,.prompt)
--expected-output <n>    Expected output tokens (default: 512)
--config <path>          Config file path (default: costguard.config.json)

Links

• Docs: https://costguardai.io/docs
• Web app: https://costguardai.io
• GitHub: https://github.com/Camj78/Cost-Guard-AI