weft-run

COMMAND, not reference: Build and run the weft program, then report what came out. Run this when the user asks for this step by name, optionally naming node ids to target. The seventeen `weft-` refere

SKILL.md

GitHub

Run this project's program and report what actually came out.

  1. Read weft-running, weft-sdp and weft-debugging (how to read what a run did). A saved example names starting parameters to run again. Node targets name where the run stops, including those nodes; --before stops before them. With no selection, run the ordinary graph roots; triggers require an explicit fire or emitted outputs.
  2. Validate before building: weft validate --file src/main.weft < src/main.weft. Fix structural errors and inspect runtime findings. Pick an existing connection with weft connect --node <id> --grant <grant>; when credentials are missing, tell the user which node needs its Connect button or weft connect in their terminal.
  3. Run weft run --detach, adding the requested selection, or weft run <example> --detach for saved parameters. Run builds automatically. For isolated inputs use --from '<node>={"port":value}', or --group '<group>={"port":value}' for a whole group. To fire one trigger, prepare with weft bake, then use --fire '<trigger>=<wake-json>'. Read the interface before choosing payloads.
  4. Wait on it with weft wait <run-id> --waiting-for-input --timeout <its cap from weft-running>; its exit code says how the run ended (the table is in weft-running). Read weft logs <run-id> for failures and weft events <run-id> --node <id> --full for values. If suspended, inspect the question or timer and report what it needs. Historical answers are evidence for deciding a new answer.
  5. When rerunning a frozen example, leave out --seed to review current computation, then weft diff <run-id> example:<name> --full. Judge the result against the user's intent. A difference is neither an automatic failure nor permission to replace the accepted example. Freeze again only after accepting the new result.
  6. Report status, the requested outcomes and any failure with its node and inputs. Name the stage this run exercised and the next stage to grow. If you start calling a completed run proof of quality, write "Wait. Read the result." and inspect the values.