task-tree-grilllisted

# Task Tree Grill Use this skill to establish or repair `task-tree.md` before execution. The output is not a prose summary; it is a visible task graph that a human can inspect, edit, and use to steer the next agent turn. `task-tree.md` is the authority. If chat history, existing files, logs, or generated artifacts disagree with the current tree, treat them as evidence only. Do not skip a node because an artifact already exists. ## Core Loop Interview the user one question at a time. Do not ask a batch of questions. For every question: 1. State the concrete uncertainty. 2. Give your recommended answer. 3. Explain what changes in the tree if the user accepts it. 4. Wait for the user before moving to the next unresolved branch. If the answer can be discovered from files, inspect the files instead of asking. Use questions only for decisions that change a node, edge, field, `GraphState`, or skill selection. ## Workflow 1. Read `task-tree.md` at the project root if it exists when you need focus; **before any write** to the tree, follow the Edit-Tree Gate in `llm-task-tree/AGENTS.task-tree.md` (Read this skill + `references/schema-template.md`, backup to `versions/`). 2. If the tree was restored or conflicts with files, follow [rollback-drift.md](references/rollback-drift.md). 3. If the user is starting or reframing a broad task, follow [interview-playbook.md](references/interview-playbook.md). 4. Write or update the smallest useful graph using [schema-template.md](referenc