Work

When Agent Plumbing Fails Quietly: A Retrospective Analysis of Open-Source Contributions to LLM Tooling and Agent Infrastructure

Year: 2026

Type: preprint

Abstract: Retrospective analysis of 11 open-source pull requests to LLM tooling and agent infrastructure (claude-mem, AnythingLLM, headroom, crawl4ai), extracted from the GitHub API on 28 October 2026. Eight merged, one open, two closed unmerged. The merged work fixes silent data loss (about 19,000 rows unindexed), approval screens that hid the content under approval, error labels that parked a provider for 30 minutes over a 60-second rate limit, and credential lookups that could never succeed on Linux. The paper's claim: agent infrastructure fails by quiet overstatement, and the fix is to make hidden states visible.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.23741621

Author: Malikian, Raphael T.

Language: English

Topic: Software Engineering Research

Subfield: Information Systems

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7224127022

Cites:

Cited by:

Related to: 10