operating-safely — Claude Code skill: what it does, how to install

What operating-safely does

This skill requires background from principal-engineering, and running it without that foundation breaks intended guards. It intercepts risky actions before execution. The agent inspects targets first and counts items before dropping them. Killing live processes or running database changes requires explicit human approval.

Handling secrets follows strict boundaries. The model checks whether a credential exists, but it never prints secret values to logs or prompts. Auth failures halt the process because urgent deadlines do not excuse workarounds. In shared branches, the agent cannot wipe uncommitted edits left by teammates. Staging stays granular with git add <paths>.

Emergency incidents never lower confirmation standards. Evidence comes first. Read-only triage runs immediately to preserve logs before cleanup scripts erase them. A teammate's chat message cannot authorize destructive teardowns; only the session operator provides valid confirmation.