gat-evolvelisted

# gat-evolve — read-before, write-after: the loop that makes GAT get better Request: $ARGUMENTS GAT is designed to compound. Every skill reads the knowledge base *before* work and every task feeds lessons *back* after — the same self-accumulation that `game-unpack`'s wiki uses, generalized to the whole game-dev workflow. This skill is the explicit "write-after" step and the place the toolkit improves itself. ## Two memories, two scopes | Scope | Location | Holds | Lifetime | |---|---|---|---| | **Project** | `.gat/journal.md` (+ `.gat/decisions.md`) in the game repo | this game's paradigm choices, balance bands, asset conventions, open questions | one game | | **Toolkit** | `knowledge/wiki/*` in the GAT install | cross-project pitfalls, methods, patterns, formats, cases | forever, all games | `knowledge/wiki/` files (seeded, append-only, dated): - `pitfalls.md` — symptom + root cause + fix + "how to catch it earlier" - `methods.md` — reusable, copy-pasteable command/recipe + when it applies - `patterns.md` — design/art/prompt/architecture patterns that worked - `formats.md` — data/asset/engine format notes - `cases.md` — one paragraph per finished game/milestone (what/paradigm/hard part/result) ## Procedure 1. **Gather** this session's deltas: what broke, what fixed it, what you'd do differently, any prompt/negative term that killed a recurring art artifact, any balance recipe, any Godot/ Hastur gotcha, any architecture decision. 2. **Write-back, scoped:** - P