The Human Prompt as an Operating System for Agentic Development

Agentic coding is often described as a model capability story: the system reads instructions, calls tools, edits files, and reports results. A Longitudinal Corpus of Human-Codex Software Work and its follow-on interpretation manuscripts by A.G. Mauro and C.A. Harris suggest a more sociotechnical object. The human prompt is not only an input string. Across repeated work, it becomes an operating surface for goals, scope, proof, privacy, authority, and completion. This paper studies that operating surface without publishing raw prompt text. It uses the approved prompt-language sample policy for The Human Prompt as an Operating System for Agentic Development, which selects an aggregate-only prompt-codebook route and keeps raw excerpts held by default. The corpus anchor is the A Longitudinal Corpus of Human-Codex Software Work aggregate snapshot cutoff 2026-07-21T07:19:35.322Z: 7,514 session files, 22,923 user messages, 79,110 assistant messages, 245,480 function calls, 238,745 shell command calls, and zero JSON parse errors. The contribution is a codebook and claim boundary for studying human operator language as workflow infrastructure. It supports a final local HCI and methods claim: repeated user instructions can be labeled as operating functions that shape agent work. It does not support public transcript release, universal human behavior claims, productivity claims, causal claims, model comparison, or public-corpus claims.

Paper: 18
Authors: A.G. Mauro and C.A. Harris
Date: 2026-07-22
Collection: Standing Framework Research

Abstract

Agentic coding is often described as a model capability story: the system reads instructions, calls tools, edits files, and reports results. A Longitudinal Corpus of Human-Codex Software Work and its follow-on interpretation manuscripts by A.G. Mauro and C.A. Harris suggest a more sociotechnical object. The human prompt is not only an input string. Across repeated work, it becomes an operating surface for goals, scope, proof, privacy, authority, and completion. This paper studies that operating surface without publishing raw prompt text. It uses the approved prompt-language sample policy for The Human Prompt as an Operating System for Agentic Development, which selects an aggregate-only prompt-codebook route and keeps raw excerpts held by default. The corpus anchor is the A Longitudinal Corpus of Human-Codex Software Work aggregate snapshot cutoff 2026-07-21T07:19:35.322Z: 7,514 session files, 22,923 user messages, 79,110 assistant messages, 245,480 function calls, 238,745 shell command calls, and zero JSON parse errors. The contribution is a codebook and claim boundary for studying human operator language as workflow infrastructure. It supports a final local HCI and methods claim: repeated user instructions can be labeled as operating functions that shape agent work. It does not support public transcript release, universal human behavior claims, productivity claims, causal claims, model comparison, or public-corpus claims.

← Back to research papers