---
name: trellis-start
description: "Initialize or resume non-trivial work in this Trellis project. Skip pure consultation and small isolated edits unless task tracking is requested."
---

# Start Session

1. Classify the request before loading context. Pure consultation and small isolated edits do not need a task or a task-creation question. Preserve risk boundaries in AGENTS.md.
2. For tracked work, verify `python3 .trellis/scripts/task.py current --source` from the actual execution worktree when entering or switching scope/worktrees. Match it to the intended task; do not treat a previously injected task as proof. Reuse the result while scope and cwd remain unchanged. Do not enumerate unrelated journals and historical tasks.
3. Read the workflow Phase Index and current step on demand. Read the current PRD, then design and implementation notes if needed. Reuse unchanged material already loaded in this session.
4. Prefer the current or directly related existing task; add a scoped repair stage rather than creating a directory for every fix. Only seek new-task consent when independently managed work has no suitable owner. Follow project-governance.md → Task Reuse. With a planning task, complete missing decisions; an approved unchanged plan needs no repeat approval.
5. For implementation load trellis-before-dev; for verification load trellis-check. Load spec indexes only for affected layers and follow relevant links, not every guideline.

The project workflow and task status are authoritative over legacy user-level Trellis skills. Missing optional tools require a supported fallback or a precise limitation report, not repeated permission requests.

For an already approved task, use official `task.py start <task>` on mismatch, then read back current and phase before implementation. Follow `.trellis/spec/hyb-oa/project-governance.md` → Active Task And Worktree Alignment; never create a shared pointer or infer hook delivery from start's success message.