implementlisted

# /implement — build it, one step at a time Takes a plan (or a brief, or a sentence) and builds it here, in this conversation, in order. It does **not** spawn a fleet of agents. Work happens sequentially in one working tree, which is both cheaper and easier to follow — you can watch every edit and stop at any point. **Progress is written into the task file as it goes.** If the session ends — you close it, or you hit a usage limit — `/implement <same file> --continue` picks up at the first unfinished step. Nothing is rebuilt. ## Arguments - **A plan or brief** — `/implement .tasks/individual-plan-export.md`. Preferred: a file from `/plan` already has criteria and ordered steps. - **A sentence** — `/implement add a CSV option to the student export`. Plan it inline first (step 2). - **Nothing** — ask what to build. Never fall back to a leftover file; building the wrong task is worse than asking. - `--continue` — resume, skipping steps already marked done. - `--step <n>` — run one step only, then stop. **Telling a file from a sentence:** strip any surrounding quotes, then check whether what remains resolves to an existing file. It does → a task file. It does not → a request in prose. A path that was meant to be a file but does not exist must **stop with "no such file"** — never fall through to treating it as a sentence and building something invented. That mistake is expensive here, because this skill writes code. Quotes are optional; arguments are not shell-parsed.