sdlc-research

# sdlc-research The **Research** phase executor — the step between restating a goal and planning it. A goal states an intention; this grounds it in the code: what the change will actually touch, what debt already lives there, and how much ceremony it earns. ## Executor resolution (host-aware) No `superpowers` / `code-review` companion covers blast-radius research, so this is **always LoopSmith's own** (same as `sdlc-retro`). Pure `grep` / `git` / markdown discipline — it degrades gracefully on any host. **Read-only.** It never edits code, plans, or standing docs. The dossier and the questions it raises are the only outputs — except the one-line `lane:` write-back in step 5. ## Proportion first Research is a cost, not a virtue. Before starting, ask whether this goal earns it: a typo fix, a rename, or a one-file change does **not** — record a one-line radius in the goal's journey note and go straight to Plan. Run the full pass when the goal touches a contract, crosses a module boundary, changes shared state, or you cannot yet name every file it will touch. Skipping it is a legitimate outcome; say so explicitly rather than producing a ceremonial dossier. ## 1. Seed Start from the goal text (`.sdlc/goals/NNNN-*.md` or the issue body) plus, if the knowledge graph is on, the `sdlc-context` brief — prior art beats re-derivation. Pull the nouns out of the goal: the modules, functions, fields, endpoints, commands, and behaviors it implies. Those are your seeds. If the goal alrea...