| name | tier-1 |
| description | The Tier 1 driver — one vertical slice, run as a single unit through the engine: a bug fix, a small feature, or a small refactor. Use when the user types /tier-1 already knowing the tier, or when route-me proposes Tier 1 and the user approves. Bug-shaped work (broken, throwing, failing, slow) routes through /diagnose first. |
Present the entry and wait for the user's explicit go before touching anything:
- the task — what's being fixed or changed, in one line;
- the tier — Tier 1, and why it is one vertical slice.
Nothing is planned or handed off until the go is given.
Satisfied when /route-me reached you — its G1 gate just presented this task and this tier and took the go. Proceed to step 1 without asking again. Typed directly, the gate is live.
Tier 1 is one vertical slice — a narrow but complete path through every layer it touches, ending in behaviour you can observe. This driver shapes the work, gets the plan approved, and hands it to the engine — it writes no code and cuts no worktree itself; /to-implement owns isolation and the build.
-
Shape the work first, if it needs it.
-
Bug-shaped — broken, throwing, failing, slow — run /diagnose before any plan. In Tier 1, diagnosis produces understanding, not commits: the plan below is the fix it points to, and the fix itself is built in the engine's worktree.
Carry its Phase 1 command into the plan as the acceptance. Diagnosis ends on one red-capable, deterministic, fast, agent-runnable command it has already run — the same four properties the brief's Verify gate demands. Naming it in the plan is what stops the brief-writer deriving it again from nothing; the brief-writer still runs it itself in the worktree, because the brief needs its own pasted output.
-
A small feature — sharpen it with /grill, the lightweight grill that writes no docs (docs are Tier 2's job). It is a conversation; nothing is branched.
-
GATE IN — the inline plan (G3). Present the plan in chat and wait for explicit approval. It lives in chat — never an artifact, never a written file.
A Tier 1 plan is a ticket that was never written down, so it carries what a ticket carries, and for the same reason: the brief-writer reads both and cannot work without these:
- Observable behaviour — what works once this lands, and how you'll observe it: the command, the screen, the request, the build.
- Layers crossed — the layers this slice cuts through, by their domain names. Names only.
- Acceptance — as end-to-end behaviour, never per layer.
- The command that proves it — for bug-shaped work,
/diagnose's Phase 1 command; otherwise the check you'd run.
Plus the base commit the worktree will branch from, named as a commit the user can see. Carrying it here is what lets /to-implement branch without a second stop: its gate asks for exactly the plan and the base, so asking twice buys nothing.
No file paths. They belong to the brief, which rebuilds them at pickup against the tree it is about to change; named here they are an anchor the brief-writer may trust instead of looking.
-
Hand off — /to-implement with input nothing: the approved chat plan is the single unit. The engine isolates the worktree and carries the rest — the brief, the build, /verified-review, /finish-branch.
When NOT Tier 1
One condition, and only one: the work turns out to be more than one vertical slice. The scout's map or the plan itself shows unclear scope (you cannot count the slices) or a blast radius wide enough that no single slice lands green. Escalate to /tier-2 rather than stretching the inline plan.
New domain vocabulary is not an escalation. It raises /domain-modeling alongside this tier — a glossary entry or an ADR committed straight to the default branch — and the slice stays Tier 1. Same for a repo with no CONTEXT.md yet.
Related: /route-me proposes the tier from evidence · /to-implement is the engine that isolates and builds · /grill sharpens a small feature · /diagnose for bug-shaped work · /tier-2 is the escalation
