goldengoallisted

# Golden Goal A goal is a contract, not a wish. The gap between "I want X" and a goal an agent can execute against is not prompt craft -- it's thinking discipline. Golden Goal helps you think through what you want, then compresses the result into a contract an agent can be judged against. The test: can someone (or something) evaluate the goal's outcome without asking you what you meant? ## Mode selection If the user provides text that looks like a goal (has an outcome, a condition, a structured prompt), start in **Sharpen** mode. If they describe intent without structure ("I want to add SSO"), start in **Compose** mode. When composing, infer answers from the codebase and conversation context. Only ask questions when the answer materially affects the goal and can't be determined from available information. A user who says "add rate limiting to the API" doesn't need to be asked "what API?" if there's one API in the repo. ## Gate -- calibrate to task weight Before running the protocol, gauge complexity: | Level | Signal | What to do | |---|---|---| | Trivial | Known fix, one file, <5 min | Skip. Just do it. | | Simple | Known outcome, small scope, low risk | Quick mode (3 lines) | | Standard | Feature, bug with unknown cause, multi-file | Full compose or sharpen | | Complex | Architecture, risky, multi-system, hard to reverse | Full + pre-mortem + risk | When in doubt, start at Standard. The gate saves effort on trivial tasks and prevents under-specification on complex