turnstile-conversate

Conversate

Process

1. Read turnstile/backlog.yml (if it exists) and the current sprint -- turnstile/sprint.yml, or a legacy root turnstile/sprint-*.yml with sprint.status: active -- to build a status snapshot.
2. Match the user's request ($ARGUMENTS plus their message) against these cases, in order:
   • Status or "what's on the board": answer directly from the snapshot (same content as /turnstile:board). No skill invocation.
   • Returning to work -- "where was I", "what was I doing", "pick up where I left off": invoke turnstile-pickup. It restores work state from the board and vault (unlike the built-in /resume, which restores a past conversation), and offers to un-park the most recently parked ticket when nothing is in progress.
   • An interrupt -- "pause this", "park this", "something urgent came up" while an item is in_progress: invoke turnstile-park with any stated reason.
   • What a piece of code does / how something is implemented: check turnstile/code/ via search_notes/read_note. If a note exists, read the source file at its aliases path and compare. Still matching: answer directly, citing the verified file. Drifted: answer from the file -- source is truth -- and dispatch brain-curator (opportunistic mode, this one file, with the purpose/exports/imports/callers you observed) to correct the note. No note: answer via Grep/Read as usual -- do not dispatch brain-curator just to backfill missing coverage; that's /turnstile:brain-init's job. No skill invocation.
   • A brand-new idea not on the board: invoke turnstile-brainstorm with the description. Exception: clearly trivial work (a typo, a tiny chore, within the quick ceiling -- quick_max_points, default 3 -- with no design question) goes to turnstile-quick instead.
   • Something broken (an error, failing test, unexpected behavior): invoke turnstile-systematic-debugger with the report. It diagnoses first, then routes the fix.
   • "Remember this" / a fact or decision the user wants recorded in the vault: invoke turnstile-remember with the user's words.
   • Cancel, kill, or drop an item: invoke turnstile-drop with that id.
   • A code review or feedback on a diff: invoke turnstile-code-reviewer.
   • Breaking an item into smaller pieces (or an epic with no children yet): invoke turnstile-breakdown with that id.
   • An existing item the user wants to move forward:
     • type: epic, or another item names it as parent: containers don't move through gates. No children yet: invoke turnstile-breakdown; otherwise report the children's statuses and route to the child at the earliest gate.
     • status: idea -> invoke turnstile-spec with that id. (An item only reaches idea after refine's design approval, so a design doc always exists -- never invoke turnstile-refine, which mints a brand-new id and cannot resume an existing item.)
     • status: ready, not in any sprint file -> invoke turnstile-sprint-plan (or turnstile-next when turnstile/config.yml says cadence: flow).
     • In the active sprint, status: in_progress, and the user says the work is finished -> invoke turnstile-review with that id. Check this before the next case.
     • In the active sprint, status: todo or in_progress (not being called finished) -> invoke turnstile-work with that id.
     • In the active sprint, status: review -> an interrupted review; if the user wants a verdict, invoke turnstile-review (it resumes). Otherwise report the status.
     • On the active board, status: parked -> invoke turnstile-pickup; it reads the resume note and offers the un-park.
     • In the active sprint, status: done -> already shipped; say so. No skill.
     • status: dropped -> cancelled; relay the recorded reason. No skill.
   • Starting a new sprint: invoke turnstile-sprint-plan (in cadence: flow it explains the mode instead of planning).
   • "What should I work on next" / pulling the next piece of work in cadence: flow: invoke turnstile-next.
   • Anything ambiguous: ask one clarifying question.