AGENTS.md vs Cursor rules vs Windsurf rules

Every comparison page on the web is a capability table someone wrote from the vendor docs in an afternoon. None of them can tell you how many real repositories ship each file, how long the median one runs, or how often two of them turn up in the same project. Those answers need a corpus, and RuleStack counts one every night.

Measured — counted from real repositories, nightlyAsserted — read from the vendor docs, dated

CompareAGENTS.md CLAUDE.md Cursor rules Copilot instructions Cline rules GEMINI.md .cursorrules Windsurf rulestwo to five

Format	AGENTS.md 2026-08-02	Cursor rules 2026-08-02	Windsurf rules 2026-08-01
Asserted — what the docs say it can express
File	AGENTS.md	.cursor/rules/*.mdc	.windsurf/rules/*.md
Status	open-standard	vendor	vendor
Cross-tool	Yes	No	No
Tools that read it	OpenAI Codex Cursor Amp Aider Zed Google Jules Gemini CLI Factory RooCode opencode Devin Warp goose VS Code GitHub Copilot coding agent Windsurf Junie (JetBrains) Kilo Code Augment Code UiPath Autopilot Semgrep Phoenix Ona	Cursor	Windsurf Cascade
Frontmatter	No	Yes	Yes
Glob targeting	No	Yes	Yes
Imports other files	No	Yes	No
Nested / per-directory	Yes	Yes	No
Multiple files	Yes	Yes	Yes
User-level scope	No	No	Yes
When it applies	Always in scope; in a monorepo the nearest AGENTS.md to the edited file wins.	Four modes set in frontmatter: Always Apply (every chat session), Apply Intelligently (the agent picks it by description), Apply to Specific Files (globs match the edited file), and Apply Manually (@-mentioned). Rules must carry the .mdc extension — a plain .md file in .cursor/rules is ignored.	Four trigger modes in frontmatter: always_on, glob, model_decision, manual. The legacy single-file .windsurfrules is still read.
Size limits	None	None	Per-rule and total character caps apply; long rules are truncated rather than rejected.
Measured — counted from real repositories on 2026-08-04
Repositories carrying it	399	179	14
Share of indexed repos	57.2%	25.6%	2.0%
Files indexed	1,373	1,280	102
Median length (words)	953	272	515
90th percentile length	2,956	1,170	1,518
Files naming commands	42.8%	29.4%	37.3%
Files containing code	36.0%	58.0%	70.6%
Median quality score	52	52	58
Median repository stars	9,402	24	119

How often they turn up together

A capability table cannot answer this at all. Of the repositories carrying the first format, this is the share that also carry the second — counted, both directions, on 2026-08-04.

Cursor rules + Windsurf rules

2026-08-04

repositories carry both. That is 6.7% of everything with Cursor rules, and 85.7% of everything with Windsurf rules.

AGENTS.md + Windsurf rules

2026-08-04

repositories carry both. That is 2.5% of everything with AGENTS.md, and 71.4% of everything with Windsurf rules.

AGENTS.md + Cursor rules

2026-08-04

repositories carry both. That is 17.0% of everything with AGENTS.md, and 38.0% of everything with Cursor rules.

What each one is for

AGENTS.md

2026-08-02

A plain-markdown README for coding agents, deliberately unopinionated: no frontmatter, no globs, no vendor keys. That minimalism is why it became the one file a dozen different agents will read, and why it carries the least per-file targeting power of any format here.

Strengths

Read by more agents than any other format — one file covers most of a team's tooling
No schema to get wrong; a plain markdown file always parses
Nested files let a monorepo scope instructions per package

Weaknesses

No glob targeting — every rule is loaded for every file, spending context
No imports, so shared rules must be duplicated across packages
No frontmatter means no machine-readable metadata for tooling to act on

Full page Browse 1,373 files

Cursor rules

2026-08-02

The most expressive format here. Many small .mdc files, each with frontmatter declaring when it should load, so a rule about migrations only enters context when a migration is open. Costs the most to maintain and only one editor reads it.

Strengths

Glob-scoped activation — the only widely-used format that keeps unrelated rules out of context
Description-based activation lets the agent pull a rule in on demand
Many small files are easier to review in a PR than one growing document

Weaknesses

Cursor-only
Frontmatter is a schema, and a malformed .mdc silently stops applying
Fragmentation: rules spread over many files are easy to duplicate and contradict

Full page Browse 1,280 files

Windsurf rules

2026-08-01

Cursor's activation model with a different vocabulary — trigger modes instead of rule types — plus hard character caps, which is the one place a format here will silently drop instructions rather than fail loudly.

Strengths

Trigger modes give the same glob and on-demand targeting as Cursor rules
Workspace and global rule scopes

Weaknesses

Windsurf-only
Character caps truncate silently — a long rule can be half-applied with no error
No imports

Full page Browse 102 files