pi-unflip
v0.1.2
Published
detects and fixes corrupted assistant text (stray CJK, Latin/Cyrillic homoglyph flips) in pi responses
Readme
pi-unflip
Fixes corrupted prose in coding-agent responses. Some providers return text with CJK characters glued into sentences ("The migration ran clean, 报告 and the index is rebuilt") and Latin letters swapped for Cyrillic look-alikes ("The сache is warm"); unflip detects this locally and, when a response looks corrupted, asks a corrector model to fix it. The corrected version replaces the original on screen and in the session itself.
Install
pi install npm:pi-unflipSo far works only with the NeuralWatt provider (
~/.pi/agent/models.jsonentry +NEURALWATT_API_KEY); without it, corrupted messages keep the original and warn.
License
MIT. See LICENSE.
