One more thing. the goal prompt is just the loop in the middle. what comes before and after matters as much. if you get stuck on either side, the ones i reach for:
𝗯𝗿𝗮𝗶𝗻𝘀𝘁𝗼𝗿𝗺/𝗽𝗹𝗮𝗻
/ce-brainstorm (compound engineering)
/ce-plan (compound engineering)
/brainstorm (superpowers)
/office-hours (gstack)
𝗿𝗲𝘃𝗶𝗲𝘄
/ce-code-review (compound engineering)
/review (gstack)
i mix and match, whichever does a step best, run them standalone. works great.
sharing a goal prompt template i use for everything.
the prompt i give it:
𝗴𝗼𝗮𝗹: {your task / the full spec you already agreed on}. keep going until the architecture and result meet the bar, not just until it runs.
𝗮𝗳𝘁𝗲𝗿 𝗲𝘃𝗲𝗿𝘆 𝗺𝗲𝗮𝗻𝗶𝗻𝗴𝗳𝘂𝗹 𝘀𝘁𝗲𝗽: real-time test the real thing (full end-to-end, plus computer use, browser, keystrokes, whatever it needs), auto review then commit, write progress somewhere sensible in the project.
𝗳𝗶𝗻𝗶𝘀𝗵: one dedicated review pass over everything.
𝗱𝗼𝗻𝗲 = every dimension at 100%, production-grade, a real user can walk in and use it.
swap out {your task} and go. new feature, refactor, all the same.
if you're doing a heavy refactor like the one in the screenshot, spin up a worktree first. the agent runs isolated and your current branch stays clean. this one's been running almost two days and yeah i'm a little nervous haha but every diff checks out so far, just slow. slow and careful beats fast and reckless though.
𝘁𝗶𝗽: don't hardcode the progress path. most projects already have a folder for progress or plans. tell it to write somewhere sensible and it finds the right place on its own.