skills/clara/references/tutorial-cases.md

# Worked examples for Clara

Choose 3–4 distinct current workflows for the user's work. The packaged starters
are fictional inputs, never precomputed answers. Read the selected skill in full.
Do not replace an unavailable capability with a mock to finish the introduction.

Use the selected current kit from `scripts/local_courses.py` in the supported
language. Follow `../../learn-with-clara/references/prepared-courses.md`. The
kit supplies valid source and practice files; `assets/onboarding/` contains
compatibility fixtures, not a separate lesson library. Custom examples may
adapt the selected own-product kit when useful.

In the bound native working chat, prepare selected inputs:

```text
python3 <clara-root>/scripts/local_onboarding_case.py --thread-id <actual-worker-id> --workflow <selected-workflow> --token <current-token> --phase demo --source <selected-absolute-file>
```

Use `--session <id>` for later lessons and repeat `--source` for multiple files.
For practice, follow the kit and actual specialist: continue the same case when
an update depends on the preceding work, or prepare a new independent attempt
with `--phase practice` after the demonstrated result has been recorded.
The helper verifies the live handoff and creates a separate attempt below the
local tutorial marker. Clara receives a project with immutable source copies,
hashes and an output directory; its specialist owns the advisory contract and
execution. Follow the selected Clara specialist's real intake,
review and completion contract; do not pretend that preparation completes it.
Reuse the saved `tutorial_case.json` after interruption instead of duplicating work.

Inspect actual outputs and record their paths as demonstration or practice evidence.
The model evaluates the meaning; the helper only checks identity, paths and state.
A reviewed draft may be the teaching objective. Never fabricate professional review
or the user's participation. Do not publish, send, call hosted interview services,
request stamps or transmit tutorial data. The compatibility marker is understood
by existing shared receipt clients; it does not merge this product's profile with Vera.

## Local transcript import and review

The `transcribe` kit uses the actual downloaded-bundle import branch. Its JSON
source explicitly identifies authored fictional text; there is no audio and no
hosted transcription execution. Prepare a real Clara case under the teaching
attempt and import the selected bundle with the current case importer. Preserve
the raw transcript, review attribution from the source metadata, and finalize
the reviewed version through the current transcript registry helper. Retain the
source notice so a learner cannot mistake this fixture for recorded audio.

The practice imports the second bundle into the same teaching case and keeps
its two unnamed speaker labels. Do not create an advisory judgement or service
approval from either input. For a user who only wants transcript storage, the
current ordinary-folder importer is a valid independent route and needs no
invented case. New recording or hosted transcription requires a separately
chosen professional handoff; this local tutorial cannot activate it.