tmp-use

# Use waz TMP instead of guessing Waz stores **verified command shapes** (binary + tokens + live values) in TMP schemas. Agents MUST assemble commands from those schemas. ## Workflow 1. `waz_tmp_list` with the user's cwd and a short query (`cargo`, `git`, `docker`). 2. Pick the exact `command` string from the list (e.g. `cargo run`, `git commit`). 3. `waz_tmp_show` for that command to see tokens and resolved values. Pass `file` (and `line`) when the user has a focused file so cargo bin/example can prefill. 4. `waz_tmp_build` with `--set name=value` pairs (or `waz_resolve` for natural language). Same `file`/`line`. 5. Return the `argv` string. Do not invent flags that were not in the schema. If the tool is missing from the list, call `waz_generate` and **wait until it finishes** (`wait` defaults to true). Then retry list/show. Do not call `waz_tmp_list` immediately after a background generate (`wait=false`) — poll `waz_generate_status` with `wait=true` (or `waz generate --wait --job <id>`) until `done` or `error`. Do not hand-write a schema. ## Rules - Never guess `--flag` names for a tool that has a TMP schema. - Never put positionals into the command name; they are tokens. - Prefer resolver-backed values (bins, branches, scripts) from `waz_tmp_show`. - If `waz_tmp_build` errors, fix tokens and retry — do not fall back to a guessed command.