CLI commands

Global options

Many commands accept:

Option	Description
-d, --dir <path>	Workspace directory (default: current directory)
-c, --config <path>	Path to shield.yaml

shieldedshell init

Create a default shield.yaml in the current directory.

shieldedshell init
shieldedshell init --force   # overwrite existing

shieldedshell doctor

Print environment diagnostics: Node version, config path, engine readiness.

shieldedshell doctor

shieldedshell run

Run a command inside the sandboxed workspace.

shieldedshell run node ./script.mjs
shieldedshell --dir ./my-project run npm test

Use -- before arguments if flags would confuse the parser.

shieldedshell shell

Interactive shell with workspace binding and overlay.

shieldedshell shell

shieldedshell verify

Static safety checks without running agents.

shieldedshell verify --type ledger --balances '...' --transfers '...'
shieldedshell verify --type routing --policies '...' --routes '...'

shieldedshell loop

Dual-agent loop with engine dispatch and prompt templates.

shieldedshell loop --engine cline --benchmark 02_ledger_consensus --dir .
shieldedshell loop --engine cursor --dir .

Option	Description
-e, --engine <name>	Agent engine (required)
--benchmark <name>	Benchmark folder under ./benchmark
--target <path>	Merge target file (default: auth_service.js)

shieldedshell orchestrate

Dual-agent loop with explicit dev/audit shell commands.

shieldedshell orchestrate \
  --dev "node ./agents/dev.mjs" \
  --audit "node ./agents/audit.mjs" \
  --dir .

shieldedshell reconcile

Run the reconciler gate on current JSON buffers (without spawning agents).

shieldedshell reconcile --dir .

@shieldedshell/core (library)

Programmatic APIs for sandbox, policy, reconcile, and orchestration:

import {
  runCommandSync,
  orchestrateDualAgentLoop,
  buildLoopCommands,
  loadConfig,
} from '@shieldedshell/core';

See the GitHub repo for exports.