Predicting Program Exit Code with LLMs and Programming Language Semantics
-
Updated
Sep 29, 2026 - Python
Predicting Program Exit Code with LLMs and Programming Language Semantics
Your agent's shell says exit 0 while a command in the chain failed. A drop-in supervised shell (Rust, embeds brush) that reports per-command and per-pipeline-stage status as JSON.
An exit code cannot tell you whether anything happened. Did it RUN, did it FAIL, and was the failure the RIGHT one?
dsh plugin: tells the agent when a non-zero exit code is an answer (grep found nothing, diff differs, test false), not a failure, so it stops re-checking commands that worked.
A Claude Code skill and hook that catches silent tool failures where exit code 0 masks a real error — it verifies a command's actual output (pytest under xdist, linters in report mode, crashed background jobs, placeholder writes) instead of trusting the exit code, and loops back to fix them.
Pone a examen los guardarraíles de una IA: comprueba que cuando detectan algo, de verdad lo bloquean, y lo demuestra mutándolos.
To associate your repository with the exit-code topic, visit your repo's landing page and select "manage topics."