Fix eval prompt collision with model flag

[KevinMeisel]

Summary

• Prevent FastAgent from consuming upskill CLI args by setting parse_cli_args=False.
• Normalize provider-prefixed model IDs when --provider is used.

Root Cause

FastAgent parses CLI args and treats -m/--message as a user prompt. Upskill’s -m model option collides with this, so the model string is injected into the chat request.

Changes

• src/upskill/cli.py: set parse_cli_args=False on FastAgent
• src/upskill/cli.py: provider/model normalization helper

Files

• src/upskill/cli.py

[KevinMeisel]

Fixes #24

[sysradium]

Oh yeah, I had that problem as well. Took me some time to figure out white -m "qwen3" just silently fails.

[evalstate]

Fixed in the latest version.