Preregistered protocol: does an automatic "error bell" prevent already-known errors in a single-user, multi-agent LLM coding team?
Preregistered protocol for a 7-day trial (planned 20-27 November 2026) of an automatic, zero-cost "error bell": a hook that recognises, in the output of a command, an error already described in a written rule and immediately shows the known fix to an LLM coding agent (Claude Code). Hypotheses, measures and analysis were fixed before the trial; the original Italian text was registered internally on 17 November 2026 (SHA-256 beb7376e0ea26731e1516ef7276dcfcff3b6cce74c557326535cf4c147663e47). The document contains an English translation and the registered Italian text. Drafted with the assistance of an AI agent (Claude, Anthropic) at the author's request; approved by the author.
Authors
- Giuseppe Calì (ORCID: https://orcid.org/0000-0002-1252-0797)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24377571
- Primary Topic
- Software Engineering Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00