---
source: https://iinventedlight.com/library/notes/ao-rules-context-files-in-practice.md
title: "What developers put in agent context files"
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
source_license: "CC BY 4.0; restated in our own words"
attribution: "Summary by iinventedlight (AI-assisted). Sources: arXiv, [2511.12884] Agent READMEs: An Empirical Study of Context Files for Agentic Coding, CC BY 4.0; arXiv, [2509.14744] On the Use of Agentic Coding Manifests: An Empirical Study of Claude Code, CC BY 4.0."
author: "iinventedlight (synthesis of cited sources); text drafted with AI assistance"
date: 2026-11-04
status: published
checked_on: 2026-11-04
as_of: 2026-11-04
confidence: medium
untested: "No experiment was rerun; only the fetched arXiv abstract pages were read. Results are as reported by the authors."
category: agent-operations
topic_area: "agent operations: rules files"
tags: agents, coding-agents, context-files, empirical
sources:
  - id: S1
    title: "[2511.12884] Agent READMEs: An Empirical Study of Context Files for Agentic Coding"
    publisher: arXiv
    url: https://arxiv.org/abs/2511.12884
    accessed: 2026-11-04
    license: "CC BY 4.0"
  - id: S2
    title: "[2509.14744] On the Use of Agentic Coding Manifests: An Empirical Study of Claude Code"
    publisher: arXiv
    url: https://arxiv.org/abs/2509.14744
    accessed: 2026-11-04
    license: "CC BY 4.0"
---
# What developers put in agent context files

## Short answer
Mostly build and test commands, implementation notes and architecture; rarely security or performance. The files grow like config code through many small edits. [S1][S2]

## Numbers (as reported)
A study of 2,303 files found tests 75.9%, implementation details 70.8%, architecture 68.1%, security 14.8%, performance 14.5%. [S1] A study of 253 manifests found shallow structure: one main heading and a few subsections. [S2]

## Takeaway
Add explicit security and performance guardrails if they matter.

## Claims
- id: C1
  claim: 2,303 context files were studied.
  source: S1
  evidence: "empirical study of 2,303 agent context files from 1,925 repositories"
- id: C2
  claim: Files evolve like configuration code.
  source: S1
  evidence: "evolve like configuration code through frequent, small additions"
- id: C3
  claim: Security requirements are rarely specified.
  source: S1
  evidence: "non-functional requirements such as security (14.8%)"
- id: C4
  claim: Manifests have shallow hierarchies.
  source: S2
  evidence: "manifests typically have shallow hierarchies with one main heading"

## Sources
- [S1] [2511.12884] Agent READMEs: An Empirical Study of Context Files for Agentic Coding. arXiv. https://arxiv.org/abs/2511.12884 . License: CC BY 4.0. Accessed 2026-11-04.
- [S2] [2509.14744] On the Use of Agentic Coding Manifests: An Empirical Study of Claude Code. arXiv. https://arxiv.org/abs/2509.14744 . License: CC BY 4.0. Accessed 2026-11-04.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted).