Gravitas 1.1: skill update targeting Gemini 4 Argon reasoning defects
The author updated the Gravitas skill after finding 95 reasoning defects across two weeks of Antigravity and Gemini 4 Argon use, adding a narration-check script and a Linux installer.
Gravitas got big update based on my reasons traces and the problem's I found during my last two weeks using Antigravity and Gemini 4 Argon. Update the skill : New in 1.1: Rules drawn from an audit of real Antigravity conversations. Of 95 reasoning defects found, 37 came from 15 runs with Gravitas loaded, in four patterns: checks described as done that never ran, frames sampled from a video read as proof of motion, gates passed by reshaping the work to fit the number, and a harness status taken as success. The always on rule's floor now answers each one, in all four languages, and three modules gain sections built from those runs. narration_check.py, a script that lists every sentence in an answer that says a check was done and names no command, file, capture or URL. On the reasoning of one audited run it flags 7 such sentences. The visual critic says first which questions an image cannot answer, so a contact sheet is never evidence of a seamless cut. Install Gravitas (Linux).sh, the native Linux installer people asked for. https://github.com/Emily2040/gravitas