| name | self-check |
| description | Exercise every WIDI tool in one fixed flow (ls, find, grep, read, ask_human, write, edit, bash) to smoke-test the harness end to end, cleaning up all artifacts. |
Run this fixed procedure in order. The goal is to trigger every tool's real execution path — not to inspect configuration. Keep each report to one or two lines. The scratch file is .widi-self-check.tmp in the current working directory; nothing else may be created.
ls the current directory; report the entry count.
find for *.md under .widi; report the hit count and names.
grep for defaultProvider in .widi/settings.json; report the matching line.
read .widi/settings.json; report what you actually read back (default provider, model, profile).
ask_human with kind=select and allowFreeInput: ask what to write into the scratch file. Options must include your own suggestion, e.g. widi self-check scratch and self-check <today's date>; the human may also type a free answer. If the request is dismissed, use your first suggestion.
write the chosen content to .widi-self-check.tmp; report the path.
edit the file: replace the word scratch (or the first line) with edited; report the change.
read the file back; report the final content.
bash a read-only check (wc -c .widi-self-check.tmp); report the output.
bash rm -f .widi-self-check.tmp, then confirm with bash (test ! -e .widi-self-check.tmp && echo gone) that nothing is left behind.
Rules:
- Never skip step 10, even if an earlier step failed — report the failure, then clean up anyway.
- If
.widi-self-check.tmp already exists before step 6, stop and tell the user instead of overwriting it.
- Do not create or modify any other file.
- Finish with a compact summary: one line per tool exercised (ok / failed), the human's answer from step 5, and any diagnostic you noticed along the way.