Internal Tooling

Build automation as productized CLI commands, not one-off scripts.

Core Rule (Summary)

If a task is likely to be repeated, shared, or non-trivial, implement it in the agent's CLI:

• Agent Jake runs tools via uv run jake ...
• Agent OpenClaw runs tools via uv run openclaw ...

The agent may still use shell/exec for command execution. The rule is about where reusable automation logic lives.

Canonical decisioning lives in references/decision-gate.md. Use this file as the source of truth for CLI-vs-one-off decisions.

Progressive Disclosure

Read only what you need: