Atbash CLI

Agent-agnostic safety judgment CLI for Agentic SPM.

Submit any AI agent action for on-chain safety judgment and get an instant verdict — PASS, HOLD, or BLOCK — powered by AI inference.

Works with any AI framework (LangChain, CrewAI, AutoGen, custom agents, etc.). No wallet or browser required — just an agent private key.

Install

npm i -g @agentic-spm/atbash

Or run without installing:

npx @agentic-spm/atbash judge "Transfer $5000 to 0xabc"

Quick Start

# 1. Generate an agent keypair (auto-saved)
atbash keygen

# 2. Set a safety policy
atbash policy set --preset strict

# 3. Submit an action for safety judgment
atbash judge "Transfer $5000 to external wallet 0xabc"

# 4. Check your agent's on-chain info
atbash whoami

Commands

atbash judge <action>

Submit an action for AI safety judgment. Submits to the blockchain, waits for AI inference, and returns the verdict.

atbash judge "Send email to 500 recipients with attachment"
atbash judge "Execute rm -rf /var/log" --context "Cleanup scheduled task"
atbash judge "Transfer $50 to vendor" --json

Exit codes: 0 = PASS/HOLD, 2 = BLOCK (useful in CI/CD pipelines).

| Flag | Description | |------|-------------| | -c, --context <text> | Additional context for the judgment | | -k, --agent-key <key> | Agent private key (if not saved) | | --timeout <ms> | Max wait time (default: 60000) | | --json | Output as JSON |

atbash policy

View and manage your agent's safety policy.

atbash policy show                      # See your current policy
atbash policy presets                   # List available presets
atbash policy set --preset strict       # Apply a preset (strict/moderate/minimal)
atbash policy set --file my-rules.txt   # Upload a custom policy file

Presets:

| Preset | Rules | Description | |--------|-------|-------------| | strict | 16 | Blocks most sensitive actions, holds cross-border and bulk ops | | moderate | 10 | Blocks critical threats, holds medium-risk actions for review | | minimal | 5 | Only blocks the most dangerous actions |

atbash whoami

Show your agent's on-chain info — name, org, risk level, plan, and jail status.

atbash whoami

atbash history

Show recent safety judgments from the chain.

atbash history
atbash history -n 20

atbash stats

Show chain-wide safety statistics.

atbash stats

atbash status <judgment-id>

Check the status of a specific judgment by ID.

atbash status j-4a7c4b8e-4b1f-4cef-9492-70b5fde61788

atbash keygen

Generate a new agent keypair. Automatically saves the private key to config.

atbash keygen                  # Generate + save to config
atbash keygen -o keys.txt      # Also export to file
atbash keygen --no-save        # Print only, don't save

atbash set <private-key>

Save an existing agent private key to config.

atbash set <your-64-char-hex-private-key>

atbash wipe

Clear all saved configuration.

atbash config

Show current configuration.

Configuration

Atbash reads config with this priority: CLI flags > environment variables > config file.

| Source | Agent Key | Node URLs | Blockchain RID | |--------|-----------|-----------|----------------| | Flag | --agent-key <key> | --node-urls <urls> | --blockchain-rid <rid> | | Env var | ATBASH_AGENT_KEY | ATBASH_NODE_URLS | ATBASH_BLOCKCHAIN_RID | | Config file | ~/.atbashrc.json | ~/.atbashrc.json | ~/.atbashrc.json |

How It Works

1. Your action is submitted as a signed transaction to the blockchain
2. On-chain AI inference (Qwen model via HybridCompute) evaluates the action against your agent's safety policy
3. The verdict (GREEN/YELLOW/RED) is stored immutably on-chain
4. The CLI polls and returns the result

All verdicts are tamper-proof and verifiable on the blockchain.

License

MIT