adrlisted

# ADR-Driven Development (front-end for `adr → sdd`) Turn an intent into one **grounded, cited, build-ready ADR**, then hand it to the deterministic `subagent-driven-development` (SDD) loop. This collapses `brainstorm → spec → writing-plans → sdd` into **`adr → sdd`** for "I know what I want, build it" work. The arc: ``` intent → GROUND → ADR (you approve) → SDD loop → you ratify ``` **Stay thin.** This skill is a four-phase orchestrator — prose plus the two embedded blocks below. The determinism lives in `sdd.mjs`, not here. One ADR doc, not a multi-file apparatus. For exploratory "not sure what I want yet" work use `brainstorming` first; for visual planning/recaps use `visual-plan` (which keeps the visual niche — `adr` carries its own build-oriented template). ## Phase 1 — Ground (scaled) Read the real code and current external practice the change touches. **Default inline; escalate to a `deep-dive` fan-out only when the change is novel, cross-cutting, or the user asks** — a deep-dive on a one-line change is the ceremony to avoid. - **State grounding:** LSP (symbols/types/refs), `git` history, an `Explore` agent for breadth. - **Research grounding:** `context7` for library docs, the cloudflare MCP for CF, a `deep-dive` fan-out for novel/cross-cutting work or on request — current external knowledge, not ~1-year-stale training. - **Record which mode was used** (inline vs deep-dive fan-out) in the ADR. **Output:** a grounding brief — every claim already cited — t