sw-session-reporter

name	sw-session-reporter
description	Finish a SolidWorks task by invoking sw-learner, honoring the saved feedback preference, asking the exact three-option consent question when required, and submitting validated feedback with retries. Use after the CAD result has been delivered for every task that involved SolidWorks modeling, API work, generated CAD code, engineering decisions, or debugging.

SolidWorks Session Reporter

Run after delivering the CAD result. Do not run for a conversation with no SolidWorks work.

Resolve the plugin root from this skill so the scripts below work regardless of the user's current working directory.

1. Build the payload

Invoke $sw-learner. Stop silently if it reports no relevant SolidWorks work. Validate the private .sw-feedback-payload.json file before any consent or submission.

2. Check the preference

Run:

python <plugin-root>/scripts/sw_session.py preference show

If the result is exactly always, skip the prompt and submit immediately. Any other result requires consent.

3. Ask consent

Ask exactly this as the final message and wait for the user's reply:

Share this session's SolidWorks knowledge with the knowledge base?

1. Yes, send now
2. Always send
3. Skip this session

Do not show the payload. Do not ask another question. Do not add closing text after the options.

4. Handle the reply

• Yes, send now: submit this payload once through the retrying utility.
• Always send: first run scripts/sw_session.py preference always, then submit this payload. This choice authorizes automatic submissions in later SolidWorks tasks until the preference is cleared.
• Skip this session: do not POST. Run scripts/sw_session.py cleanup to remove the temporary payload.

Do not interpret ambiguous replies as consent. Repeat only the same three options if necessary.

5. Submit

Run:

python <plugin-root>/scripts/submit_feedback.py .sw-feedback-payload.json

The utility validates required fields, removes empty arrays, uses quoted curl transport, retries server errors or timeouts up to three times with about two seconds between attempts, stops immediately on 4xx, and writes the returned feedback ID to .sw-learner-state.json.

After success, run scripts/sw_session.py cleanup. Report only the returned feedback ID. If the server rejects or never accepts the payload, do not expose raw server data or bother the user; leave the CAD result unchanged and clean up the temporary payload.