@tank/rule-creator

To prevent dangerous shell commands without custom code, you define enforceable rule atoms using YAML frontmatter that bind to command events with match patterns and block policies. This replaces scripting with machine-enforced governance.

Policy as code for agent safety is a method of controlling agent behavior by defining block, warn, or allow rules in structured configuration files. It uses event bindings and match logic instead of procedural scripts to enforce operational governance.

You create a tool allow-list for file-write events by authoring rule atoms that bind to file-write actions and apply allow policies based on specific match patterns. This restricts agent capabilities to explicitly permitted operations.

Yes, you can combine multiple agent safety rules into a single bundled policy set through bundle composition. This allows you to layer instruction atoms and rules together to create comprehensive governance for real projects.

You should use warn policies instead of block policies for command gating when addressing code-quality issues or style warnings, reserving block policies for strictly preventing destructive commands and high operational risks.