What problem does it solve? Harness projects need a project-wide control over how much the AI decides on its own versus asking the human first. Manually editing the decision-policy file risks mangling its structure; this Skill flips the Active mode line safely, with backups, confirmation gates, and rollback via git. ## Core Features & Use Cases - Three decision modes: Mode 1 (human decides every judgment call), Mode 2 (AI decides per the preset rubric, escalating only red lines), and Mode 3 (AI decides per your own custom rubric). - Surgical, non-destructive edits: rewrites only the single Active mode: line of .harness/rules/25-decision-policy.md, with a timestamped .bak backup and a clean-git-tree precondition so git reset is a full rollback. - Custom rubric seeding: on a first switch to Mode 3, interactively collects your decision principles and seeds the Custom rubric section of .harness/decision-rubric.md. - Use Case: You want the AI to stop interrupting you with judgment calls, so you ask to switch to Mode 2; the Skill shows the one-line diff, asks for confirmation, applies it, and reports the backup path. ## Quick Start Ask the AI to switch the project's decision mode to Mode 2 so the AI decides routine judgment calls on its own.
