find-your-unknownslisted

# find-your-unknowns A more capable agent traverses more of the codebase, so it hits more decisions you never wrote down — and drifts. The fix is to find those decisions *before* the big run. Your prompt is a **map**; the codebase is the **territory**. This skill surfaces everywhere the territory holds something your map doesn't. ## When to use vs. skip - **Use** before a large, ambiguous, or architecture-shaping task, or when you (or the user) are new to this codebase or domain. - **Skip** for a small, fully-specified change where the territory is already known — running the full battery would be ceremony. ## Inputs - A **task/goal** (what the user wants built or changed). - A **target repo or subsystem** you can read. - Optionally: reference code/mockups, and context about the user's taste, stage, and constraints. ## Workflow Don't blindly run all six. Pick by the *kind* of unknown in play (see table), run those, then consolidate. Note the **When** column: only patterns 1–4 run during this scoping pass. Pattern 5 is an instruction you *produce now* but inject into the later build run; pattern 6 runs *after* the build. Each pattern's full prompt template is in `references/prompt-patterns.md`. | Kind of unknown | Pattern | When | Skip if | |---|---|---|---| | "I don't even know what I don't know here" | **1. Blind-spot pass** | scoping | rarely — this is the anchor | | Aesthetic / "know it when I see it" | **2. Brainstorm & prototype** | scoping | the surface isn't