task-tunerlisted

# Task Tuner ## Platform compatibility When running in ChatGPT or Codex, read `../../references/codex-compatibility.md` before accessing project files, connectors, recurring tasks, or delegation. Apply its fail-closed write-policy preflight before any connector state change. A correction that only fixes today's output is a correction the owner has to repeat. This turns "the briefs are too long" into a line in a file that changes every future run. ## Contract block **What it reads.** The correction in the owner's own words, the run being corrected, the current `Task Settings` including its `## Action controls` section, and `references/action-controls.md` for what tuning may touch. **What it produces.** A concrete rule, the exact before and after of the setting, and once the owner approves, an updated `Task Settings` with a row appended to the tuning history. **What it never does.** Change a setting without showing the before and after. Write to `Boundaries` without saying explicitly that a boundary is being changed. Turn an action on. Widen a scope. Interpret a correction about one run as a permanent rule without confirming that is what the owner meant. **What needs your approval.** Every write to any of the owner's context files. ## The four steps ### 1. Translate the feeling into a rule The owner's words are the input, not the output. Restate them as a rule specific enough to apply mechanically. | The owner says | The rule | |---|---| | "The briefs are too long"