v1.2.0

What changed

Claude Code output style. The plugin now ships simple-english as an output style. Install the plugin, run /config, open Output style, and pick simple-english. The style applies STE to every reply. The skill still triggers per task. (#9)

Vocabulary fixes, verified against the Issue 9 dictionary. The consistency pass now separates free choices from words the dictionary already ruled on: check/verify/confirm/ensure become "make sure that" in strict mode, run becomes operate, execute becomes do, display/render/present become show, delete becomes erase or remove. The fall and oil rulings are corrected. Thanks to @dfch for #5 and the audit in #3 and #4. (#2, #3, #4, #5, #7)

Cross-model benchmarks. A separate Pi harness tested 4 non-Claude models (GLM-5.2, three GPT-5.6 variants) on the same 8 tasks. The skill cut measured violations 67-88% on all four. Thanks to @Whamp. (#8)

Plugin marketplace. The repo is a native Claude Code plugin marketplace. Thanks to @akolotov. (#6)

Verification

The vocabulary fix was A/B tested: two agents, same input, strict mode. The old skill picked the rejected verb "run". The fixed skill produced zero rejected words. The output style was tested headless: 0 linter violations with the style on.