What Open-Source Projects Tell Their Coding Agents: A Census of AGENTS.md and CLAUDE.md

Pawel Zarzycki

cs.SEOct 22, 2026

What Open-Source Projects Tell Their Coding Agents: A Census of AGENTS.md and CLAUDE.md

Authors: Pawel Zarzycki

Abstract

Coding agents can be guided with instruction files in a repository. We examine this emerging practice through a census of 22,854 popular, active GitHub repositories in thirteen languages, treating each repository's root 'AGENTS $.$ md' and 'CLAUDE $.$ md' as one instruction set. Agent instructions appear in 26.3% of repositories and 59.6% of the most-starred; where both files exist, 'CLAUDE $.$ md' usually points to 'AGENTS $.$ md'. We analyse 341,852 statements using model-assisted annotation and human review. The annotations reveal a practical mix of project knowledge and rules about code, building and workflow. Collaboration rules name concrete requirements, from test plans to commit formats, yet 74.0% of directives give no reason. Projects disagree on AI attribution: commit-message rules more often prohibit it than require it, while pull-request rules more often require disclosure. A third of instruction sets have a generator-associated signature, a tool-inserted block, shared text or an AI-marked creation commit; four in five are edited after creation. Prespecified tests associate repository popularity with rules on pull requests and AI contributions. Exploratory analyses link activity to process-oriented content and show that LLM and agent projects are over-represented among adopters. The models agree much more closely with each other than with the human reviewer's topic assignments, a caution for automated annotation. We release the annotations, taxonomies and pointers to the analysed file versions.

Figures & tables

Figure 1: Adoption rises with popularity, and new adopters increasingly start with AGENTS.md . (A) Share of repositories with agent instructions by star tier (census of 22,854). (B) New adopters per month by their first surviving agent file; the dashed line marks Claude Code’s support for AGENTS.md (2026-09-18); the last bar is a partial month.

Figure 2: Rules most often concern code conventions, building and agent workflow. Share of repositories whose instruction set has at least one rule (blue, 95% bootstrap intervals) and any statement (grey) in each subject area. Sonnet’s codes on 992 sample repositories, reweighted to the population.

Category	% requiring	% prohibiting
test plan / verification evidence	27.2	4.4
use and complete the PR template	24.2	1.6
linked issues and related PRs	23.3	1.3
title format (type, tag or scope prefix)	19.3	1.7
summary of the change	17.8	0.1
screenshots and visual evidence	15.6	1.1

Table 1: Selected elements of rules on pull-request descriptions and commit messages: share of the files with a rule on the subject that require or prohibit each element (Haiku, full corpus, uncorrected). All categories are in the appendix.

Figure 3: Most instruction sets contain marked rules, prohibitions and reasons somewhere, but each is a small share of their directives. Share of repositories with at least one directive of each kind (dots, 95% bootstrap intervals) and its mean share of a repository’s directives or statements (diamonds). Sonnet’s codes on the sample, reweighted.

Hypothesis	Verdict	Effect (95% CI)
H1 AI-marked share: more agent rules, permission/escalation rules and prohibitions	not supported	IRR 0.977 (0.959–0.996), 0.992 (0.963–1.021), 0.981 (0.965–0.997)
H2 Collaboration: more workflow rules, fewer descriptions	mixed (2 of 4)	merged PRs: workflow IRR 1.027 (1.009–1.046), descriptions 0.876 (0.858–0.894); human authors: workflow 0.975 (0.959–0.992, opposite), descriptions 0.993 (n.s.)
H3 Compiled languages: more build/run statements	supported	OR 1.68 (1.48–1.91)
H4 Emphasis: more rules on agent failure modes	not supported	OR 0.93 (0.88–0.99)
H5 CLAUDE.md more descriptive than AGENTS.md , same repo	supported	+6.3 pp (4.2–8.4); Wilcoxon p = 1.3e-7
H6 Stars: more pull-request or AI-policy rules	supported	OR 1.36 (1.29–1.44)

Table 2: Prespecified hypotheses: verdicts and principal effects. Ratios are per standard deviation of continuous predictors; H3 compares compiled with other languages, and H5 is a paired percentage-point difference. Benjamini–Hochberg correction covers 11 tests. 5,587 repositories; H5 uses 484 pairs.

Figure 4: Activity goes with process- and agent-oriented content, and AI projects are over-represented among adopters. (A) Exploratory models: odds ratio of covering each area per standard deviation of

\log(1+\text{commits in the past year})

, with other predictors held fixed (blue: significant after Benjamini–Hochberg, grey: not; Haiku’s codes, 5,586 instruction sets). (B) Adopters against 4,699 matched repositories without agent files, by project type, within star tier and language; the square adds controls for age and activity and excludes agent-tooling projects. Project types are model-classified and indicative.

Appendix figures & tables18 assets

Supplementary material from the paper’s appendix.

Appendix

Check	Result (n = 60)
Statement cut correctly	55 (92%; Wilson 95% CI 82–96%); 5 “part missing”
Restatement faithful	56 (93%; 84–97%); 4 “loses meaning”
Listed parts (elements) right	56 (93%); 4 “some missing”
Free-text subject fits	57 (95%)

Appendix

Table 3: Human review of 60 statements. Top: checks of extracted statements. Bottom: agreement of the blind reading with Haiku, Sonnet and Opus (ranges across models).

Language	Repos	AGENTS.md	CLAUDE.md	Either
TypeScript	4,048	34.4%	24.5%	40.8%
Rust	1,706	28.1%	18.9%	32.7%
Go	2,264	24.5%	16.6%	28.9%
Swift	474	23.4%	17.9%	28.1%
Kotlin	611	21.1%	14.9%	26.0%
Java	1,262	20.9%	12.4%	24.6%

Appendix

Table 4: Repositories with each agent file, by primary language. Census of 22,854 repositories.

Format	Repos	Share	Also has /AGENTS.md
/AGENTS.md	4,954	21.7%	—
/CLAUDE.md	3,440	15.1%	69.4%
.github/copilot-instructions.md	1,024	4.5%	52.3%
.cursor/rules/ ( $\geq$ 1 .md/.mdc)	347	1.5%	64.6%
.github/instructions/	273	1.2%	59.0%
/GEMINI.md	179	0.8%	83.2%

Appendix

Table 5: Agent instruction formats in the 22,854 repositories, listed one level deep. The last column is the share of repositories with the format that also have a root AGENTS.md .

Relationship	Repos	Share
AGENTS.md only	2,549	42.4%
AGENTS.md only (symlink to another file)	17	0.3%
CLAUDE.md only	1,040	17.3%
CLAUDE.md only (symlink to another file)	12	0.2%
both: CLAUDE.md is a symlink to AGENTS.md	629	10.5%
both: AGENTS.md is a symlink to CLAUDE.md	155	2.6%

Appendix

Table 6: How the two files relate in the 6,006 repositories with an agent file. The rules are applied in order ( Section 3.1 ).

Period	New adopters	First file AGENTS.md	First file CLAUDE.md	Both at once
2025-01 – 2025-08	536	23.9%	73.7%	2.4%
2025-09 – 2025-12	759	54.2%	36.8%	9.1%
2026-01 – 2026-04	2,178	48.3%	35.9%	15.8%
2026-05 – 2026-10-16	2,532	60.0%	17.1%	22.9%

Appendix

Table 7: New adopters by the date of their first surviving agent file, and which file came first.

Trace	Repos	Share
generator signature ( claude /init line or variant, Codex /init headings)	948 of 5,998	15.8%
of which claude /init ’s opening line, verbatim	692	11.5%
tool-inserted block (Next.js, Turborepo, Nx, Laravel Boost, beads)	93	1.6%
paragraph shared verbatim with files of $\geq$ 2 other owners	205	3.4%
of which with no generator signature or tool block	104	1.7%
whole file near-duplicate of another owner’s file	60	1.0%

Appendix

Table 8: Traces of origin per instruction set. Shares of all instruction sets for text traces and of those with a file history for history traces. Traces identify signatures, shared text or marked commits; they do not establish authorship.

Type	Ours, raw	Ours, corrected	Reference	Verdict
AI Integration	72.5%	71% (59–92)	22% (17–28)	more common in ours
Documentation and References	83.7%	72% (62–85)	29% (24–36)	more common in ours
Development Process	78.0%	91% (82–100)	68% (62–74)	more common in ours
DevOps	38.9%	29% (23–36)	17% (12–22)	more common in ours
Security	41.8%	28% (22–33)	17% (13–23)	more common in ours
UI/UX	24.3%	20% (14–32)	6% (4–10)	more common in ours

Appendix

Table 9: Share of files with each instruction type of Chatlatanagulchai et al. [2026] : ours raw, ours corrected for over-labelling with the Rogan–Gladen estimator (95% intervals), and the reference study’s human coding of its AGENTS.md and CLAUDE.md files. The samples differ in frame and date, so a difference shows that the populations differ, not why.

Derived type	Adopters % (n = 6,006)	Controls % (n = 5,920)
data / ML / AI (any AI focus)	32.2	18.6
library / framework	23.0	32.0
application	22.3	22.3
CLI / developer tool	10.7	13.2
infrastructure / DevOps	8.0	6.0
docs / learning	2.6	6.0

Appendix

Table 10: Project types of all adopters and all classified controls (model-classified by Claude Haiku 5.5 from repository metadata; indicative, see Section 4.5 ).

Kind	% of repos with $\geq$ 1	Mean share of statements	Census (Haiku, corr.)
rule	98.2 (97.3–98.9)	52.6 (50.9–54.4)	53.0
reference	89.7 (87.8–91.6)	17.6 (16.5–18.8)	16.2
description	88.6 (86.6–90.5)	29.8 (28.3–31.3)	30.7

Appendix

Table 11: Kinds of statement per instruction set: Sonnet’s codes on 992 sample repositories, reweighted, with 95% bootstrap intervals; the last column is the corrected Haiku census.

Area	% with a rule (95% CI)	% with any statement
code conventions and quality	74.4 (71.4–77.3)	77.0
build, environment and delivery	72.6 (69.4–75.7)	87.6
agent workflow and conduct	67.3 (64.0–70.3)	68.5
documentation	61.7 (58.4–65.1)	77.3
testing	60.2 (56.8–63.4)	76.7
version control and collaboration	58.0 (54.3–61.7)	60.9

Appendix

Table 12: Subject areas: share of repositories with at least one rule (95% CI) and with any statement (the data of Fig. 2 ).

Predictor	Significant / tests
commits in 12 months	11 / 13
age	9 / 13
project type (joint)	6 / 13
language (joint)	6 / 13
human authors	4 / 13
AI-marked share	4 / 13

Appendix

Table 13: Exploratory models: tests per predictor that pass Benjamini–Hochberg within the exploratory family (99 tests). Content models use 5,586 instruction sets; the adopter-versus-control row uses 5,286 adopters and 4,699 controls.

Type	Adopters %	Controls %	Odds ratio (95% CI)	p (FDR)
data / ML / AI	32.2	18.6	2.38 (2.15–2.64)	< 0.001
LLMs / generative AI / agents (AI focus)	27.9	13.3	2.76 (2.48–3.08)	< 0.001
infrastructure	8.6	6.6	1.38 (1.17–1.61)	< 0.001
application	21.9	21.9	1.00 (0.91–1.11)	0.96
developer tool	10.9	13.6	0.77 (0.68–0.87)	< 0.001
library / framework	23.0	33.3	0.58 (0.53–0.64)	< 0.001

Appendix

Table 14: Adopters and matched controls by project type (within star tier and language strata; controls reweighted to the adopters’ strata). Types are model-classified and indicative.

Field	PI agrees with Haiku	Wilson 95% CI	Cohen’s $\kappa$ (95% CI)
AI focus	27 / 29 (93.1%)	78–98%	0.80 (0.44–1.00)
type	16 / 29 (55.2%)	38–72%	0.43 (0.23–0.63)
derived seven-way type	17 / 29 (58.6%)	41–74%	0.50 (0.30–0.70)

Appendix

Table 15: The human reviewer’s blind check of Haiku’s project types on 29 repositories.

Stars	Repos	AGENTS.md	CLAUDE.md	Either
100k+	89	52.8%	31.5%	59.6%
10k–100k	3,059	32.3%	21.5%	37.0%
5k–10k	3,087	26.1%	17.9%	31.2%
1k–5k	16,619	18.7%	13.2%	23.2%

Appendix

Table 16: Repositories with agent instructions by star tier (the data of Fig. 1 A).

Category	% requiring	% prohibiting
Test plan / verification evidence	27.2	4.4
Use and complete the PR template	24.2	1.6
Linked issue and related PR references	23.3	1.3
Title format / type, tag or scope prefix	19.3	1.7
Summary of the change and its behavior effect	17.8	0.1
Screenshots, recordings and visual evidence	15.6	1.1

Appendix

Table 17: Pull-request descriptions: share of the 1,345 files with a rule on the subject that require or prohibit each element.

Category	% requiring	% prohibiting
Conventional Commits / structured header format	36.7	1.8
Allowed commit types and when to use each	23.5	0.6
Subject mood, case and punctuation	15.1	1.3
Scope / component / area prefix	14.9	0.8
Body presence, layout and wrapping	12.6	1.2
Rationale and change description in the message	12.6	2.0

Appendix

Table 18: Commit messages: share of the 1,739 files with a rule on the subject that require or prohibit each element.

Category	% requiring	% prohibiting
Tests must accompany changes (and exemptions)	28.2	1.9
Frameworks, libraries, helpers and selectors	19.5	2.2
Test structure, granularity and conventions	17.4	2.4
Required scenarios, edge cases and error paths	16.8	0.1
Choice of test type / level	15.5	1.7
Regression test for bug fixes	14.3	0.3

Appendix

Table 19: Test writing: share of the 2,586 files with a rule on the subject that require or prohibit each element.

Category	% requiring	% prohibiting
Review process, reviewers, tools and gating	22.2	2.7
Correctness, edge cases and error handling	20.2	2.0
Context gathering and review method	17.7	1.5
Project guidelines, checklists and repo-specific checks	17.5	1.2
Evidence, verification and confidence of findings	14.5	1.5
Review report format, verdict, tone and language	14.5	1.0

Appendix

Table 20: Code review: share of the 401 files with a rule on the subject that require or prohibit each element.

AI Coding Agents

AI coding agents use language models and software tools to plan, edit, test, and maintain code with varying levels of autonomy.

See all 326 papers

Detecting AI Coding Agents in Open Source: A Validated Multi-Method Census of 180 Million Repositories
Jun 23, 2026Arsham Khosravani, Audris MockusSoftware Engineering Coding Agents
Agent or Human? Detecting Commits by AI Coding Agents in the Absence of Authorship Signatures
Oct 26, 2026Kosei Horikawa, Hao Li, Yutaro Kashiwa +5Authorship Attribution AI Coding Agents
Nara Institute of Science and Technology, Ikoma, Japan · Queen’s University, Kingston, Canada · École de technologie supérieure - Québec University, Montréal, Canada
A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities
Jul 29, 2026Wenhao Yang, Runzhi He, Minghui ZhouAI Coding Agents AI Agent Benchmarks
Peking University, Beijing, China

Jun 23, 2026J/K move · Enter open · S save

Detecting AI Coding Agents in Open Source: A Validated Multi-Method Census of 180 Million Repositories

Arsham Khosravani, Audris Mockus

Software Engineering Coding Agents Mining Software Repositories Open-Source Software Development

Open paper AskPDF arXiv

Generative AI coding agents are entering the open-source supply chain, yet their diverse and often invisible traces leave their prevalence poorly understood. We introduce a multi-layered detection framework that integrates configuration-file scanning, commit-message analysis, author-identity matching, and bot-signature lookup across World of Code (180M+ Git repositories), classifying agent traces into four behavioral types. No single method captures more than a fraction of activity: multi-method detection identifies 850,157 Claude Code commits in one snapshot, of which bot-account lookup_the signal most adoption studies rely on_recovers only 28,154 (3.3%), a 30x relative-recall gap, so single-signal prevalence estimates are biased low by at least this factor. Every detection pattern is hand-validated (495 labels) with per-cell precision and Wilson confidence intervals. Across snapshots from December 2024 to April 2026, commit-attributed agents generate over 320,000 commits per month; Claude Code leads (886,122 commits across 17,295 projects) and dominates silent, configuration-file-only adoption (21,078 projects). Compared against an independent pull-request census (AIDev), the two channels capture nearly disjoint agent populations_a PR census misses 79% of commit-detected Claude Code adopters and essentially all Codex adopters_and different kinds of work: PR-deployed cloud agents (Codex, Cursor) surface as feature work, while commit-deployed in-editor agents (Claude Code, OpenHands, Aider) surface as maintenance. The observed work profile follows deployment and detection mode rather than the tool itself, so no single channel is representative.

Abstract

AI Coding Agents

Explore similar work

Detecting AI Coding Agents in Open Source: A Validated Multi-Method Census of 180 Million Repositories

Agent or Human? Detecting Commits by AI Coding Agents in the Absence of Authorship Signatures

A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source Communities