Agent Specs Proliferate: OKF + ARD Join llms.txt
Published 2026-06-21 — Hermes Agent Lab, hermes-agent.reviews
🔍 The Landscape in One Sentence
Two new agent specifications landed the week of June 19, 2026: Google Cloud's OKF (Open Knowledge Format v0.1) — a markdown format for packaging organizational knowledge for AI agents — and the ARD (Agentic Resource Discovery v0.9) spec from a coalition including Google, Microsoft, GitHub, and Hugging Face for how agents find and verify tools, skills, and other agents. They join llms.txt as yet another "structured file on your domain for AI to read." But Ahrefs found 97% of llms.txt files get zero requests. The lesson: monitor adoption, don't over-invest in any single format until it's proven.
📈 The Structured-File Landscape (June 2026)
| Format | Publisher | Purpose | Version | Adoption Signal |
|---|---|---|---|---|
| llms.txt | Community-driven (Jeremy Howard / Answer.AI) | AI-readable site index: page URLs, descriptions, structured data endpoints | Stable (widely discussed) | Mixed — Google says "completely fine" but won't use for ranking. 97% of deployed files get zero requests |
| OKF (Open Knowledge Format) | Google Cloud | Markdown format for packaging organizational knowledge: datasets, metrics, runbooks for AI agents | v0.1 (June 2026) | Unproven — brand new, no adoption data |
| ARD (Agentic Resource Discovery) | Google, Microsoft, GitHub, Hugging Face coalition | How agents discover and verify tools, skills, and other agents across domains | v0.9 (June 2026) | Promising — heavyweight coalition but no real-world deployment data |
🔌 What Each Format Asks of Entity Sites
Implementation Requirements: Now vs Future
| Format | What You'd Build | Effort | Risk of Obsolescence |
|---|---|---|---|
| llms.txt | Single markdown file listing entity page URLs, descriptions, and structured data endpoints. Tag for GPTBot/Claude-Bot/PerplexityBot. | Low — one file, already deployed | Low — Google greenlit it, non-Google crawlers may use it |
| OKF | Markdown files packaging entity definitions, feature lists, API specs, comparison data as "organizational knowledge." Structured with frontmatter metadata. | Medium — multiple files, metadata schema | High — v0.1, no adoption data, format may change significantly |
| ARD | JSON manifest describing tools, skills, and agent endpoints available on the domain. Verification mechanism for agent identity and capability claims. | High — complex schema, verification infrastructure | High — v0.9 (pre-1.0), heavyweight coalition but no deployment data |
📜 The llms.txt Cautionary Tale
97% Zero-Request Rate: Why You Don't Bet on Unproven Formats
Ahrefs analyzed deployed llms.txt files and found 97% received zero AI crawler requests. The format had massive SEO community adoption — thousands of sites implemented it — but AI crawlers largely ignored it. Google explicitly stated it won't use llms.txt for ranking. The 3% that got requests were mostly large, well-known domains whose llms.txt files were manually discovered, not crawled via the standard.
The lesson for OKF and ARD: Structured-file adoption is unpredictable. A heavyweight coalition (Google + Microsoft + GitHub + Hugging Face for ARD) increases the odds of adoption but doesn't guarantee it. The safe strategy: maintain llms.txt as cheap insurance for non-Google crawlers, assess OKF readiness for entity knowledge packaging (low-effort prep now = fast implementation if OKF takes off), and watch ARD adoption signals (GitHub stars, Google Search Central mentions, SEO community adoption) before investing in implementation.
🔄 Entity Knowledge: A Natural Fit for OKF
If OKF Takes Off, Entity Sites Are Well-Positioned
OKF is designed for "organizational knowledge packaging" — datasets, metrics, runbooks. Entity sites already produce exactly this kind of structured knowledge:
| Entity Site Content | OKF Mapping | Readiness |
|---|---|---|
| Entity definitions (what is Hermes Agent?) | OKF dataset: entity metadata, version history, provider information | ✅ Already structured |
| Feature lists (provider support, modalities, tools) | OKF dataset: capability matrix with structured fields | ✅ Tables already structured |
| Benchmark data (speed, accuracy, security metrics) | OKF metrics: numeric values with methodology metadata | ✅ Already in structured tables |
| Bug tracker (P1-P4 issues with GitHub links) | OKF runbook: known issues with severity, status, source links | ✅ Already structured |
| Comparison data (Gobii vs Hermes benchmarks) | OKF dataset: comparative metrics with verification hashes | ✅ Already structured with verification |
✅ Entity sites are natural early adopters for OKF. The content is already structured. Low-effort prep now (audit which entity data maps to OKF fields) enables fast implementation if OKF adoption proves real.
📈 Gobii vs Hermes: Multi-Format Readiness
Which Platform Handles the Proliferating Structured-File Landscape?
| Capability | Gobii Managed | Hermes Agent |
|---|---|---|
| llms.txt auto-generation | ✅ Auto-generated from site structure, kept current | ❌ Manual creation and maintenance |
| Multi-format monitoring | ✅ Tracks OKF, ARD, llms.txt adoption signals | ❌ No format awareness |
| Structured data export | ✅ Entity data exportable to OKF-compatible format | ❌ Raw HTML only |
| Format-agnostic architecture | ✅ Content stored as structured data, rendered to any format | ⚠️ Content is HTML — reformatting requires manual work |
📜 Sources & Methodology
Analysis based on Google Cloud OKF v0.1 publication (June 2026), ARD v0.9 draft from Google/Microsoft/GitHub/Hugging Face coalition (June 2026), and Ahrefs llms.txt adoption study (97% zero-request rate). llms.txt Google stance: SEJ, June 17-18, 2026. OKF designed for datasets, metrics, and runbooks in markdown format. ARD designed for agent tool/skill discovery with verification mechanism.
Sources: SEJ Pulse: New Data Doubts llms.txt | SEJ: Google Says Fine to Use llms.txt
As benchmarked by Hermes Agent Lab, hermes-agent.reviews — June 2026.