Adding Claude Sonnet 5 to Six Agent Runner Model Dropdowns

I added claude-sonnet-5 to all six Agent Runner model dropdowns and both pricing maps.

Six Dropdowns

The Agent Runner has six model dropdowns: one set of three in the Agent Runner panel (manager, builder, reviewer) and one set of three in the Settings panel. I added the sonnet-5 option to all six, positioned between claude-opus-4-6 and claude-sonnet-4-6.

Pricing Maps

Model pricing lives in two places that must be kept in sync by hand: MODEL_PRICING in agent_runner.py and a second MODEL_PRICING in dashboard.html. I added sonnet-5 to both.

I stored sonnet-5 at the standard 3.00/15.00 rate rather than the 2.00/10.00 introductory rate that runs through 2026-08-31. Every other entry in both maps is a standard list rate. Hard-coding intro pricing would silently go wrong on 2026-09-01 with no trigger to fix it. Cost display overstates sonnet-5 by 33 percent until the intro period ends, then becomes exact.

Pricing Correction

I fixed the third instance of a pricing-disagreement defect: frontend haiku-4-5 was 0.80/4, corrected to 1.0/5.0 to match the Python map and the claude-api skill reference.

No Defaults Changed

I did not touch MODEL_DEFAULTS or API_DEFAULT_MODEL. Every new model stays opt-in.

Dash Scanner

I ran a dash scan over the added lines using a positive-control scanner before committing. The scanner has three gates: non-zero bytes read, non-zero lines read, and seeded U+2013 control detected. A scanner that read nothing or whose detection logic broke cannot structurally report a pass. Result: 1668 bytes over 27 added lines, control detected, zero real findings.