skill-runnerlisted

# Skill Runner Use this skill to run one real development task in a separate agent session while the main session stays focused on decisions, review, and synthesis. The default mindset is daily development, not benchmark tuning. Do not rerun the same task just to optimize a skill. Learn from varied real tasks, and change a skill only when a reusable workflow defect is clear. This skill may repair itself when a real `skill-runner` run reveals a reusable runner defect, such as false blocker classification, unsafe supervision logic, or brittle artifact parsing. Keep those patches small and commit them separately from product-task changes. ## Philosophy Keep skills small, composable, and boring. - Prefer principles over long procedures. - Prefer stop conditions over exhaustive branching. - Prefer one obvious path over clever options. - Prefer scripts for deterministic mechanics. - Prefer references only when optional detail is genuinely needed. - Prefer deleting or shortening stale guidance over adding more rules. - Do not encode task-specific lessons as universal rules. - Do not add a rule unless it prevents a repeated or high-severity failure. - When a better model can infer it from context, leave it out. ## Default Flow 1. Parse the user task and selected skills. 2. Choose the task workspace: the repo where the user prompt was triggered, unless the user explicitly names a different target repo. Do not switch to the custom skill source repo merely because a select