Attestation type conventions

The contract's type field is free-text. Consistent prefixes make attestations interoperable, filterable, and interpretable.

Standard prefixes

Prefix	Meaning	Example
skill:	Verified competency in a specific skill	skill:literature-review-v2
diploma:	Completion of a formal program	diploma:mit-cs-2026
license:	Authorization to operate in a regulated domain	license:medical-advice-v1
audit:	Behavioral review passed in a specific period	audit:safety-review-2026q1
identity:	Identity claim tying agent to a concrete system	identity:anthropic-claude-sonnet-4-6
badge:	Community recognition, non-verifying	badge:early-adopter
membership:	Confirmed membership in an organization	membership:acme-dao

Naming conventions within a prefix

• Lowercase, hyphens for spaces, no whitespace
• Version when meaningful: skill:literature-review-v2 makes the version explicit
• Date-stamp when period-scoped: audit:safety-2026q1 makes the review period clear
• Provider-specific identity: identity:<provider>-<model>-<version>

How to propose a new prefix

Open a GitHub Discussion describing:

1. The proposed prefix and meaning
2. Who would issue attestations of this type
3. How consumers would use them
4. An example

If consensus forms, submit a PR adding it to this table. See contributing.

Guidance for issuers

A good attestation answers three questions in its description:

1. What specifically was verified?
2. How was it verified?
3. When / under what conditions?

Weak description

Good at code review

Strong description

Passed the SecureCode-2025 benchmark at 91% (top decile). Verified 2026-03-15 against commit sha abc123 of the agent's model config. Full report: ipfs://bafkrei...

Guidance for consumers

Attestations are claims, not guarantees. Trust depends on:

1. Who issued it — check the issuer wallet and its reputation
2. What evidence it references — the uri field should point to verifiable detail
3. Whether it's been revoked — check revoked flag
4. How the issuer votes with their track record — selective issuers accumulate trust; spammy ones lose it

For high-stakes decisions, prefer attestations from a small whitelist of known reputable issuers.

Provider-specific conventions (community maintained)

As the ecosystem grows, specific providers and communities adopt conventions. Pull requests welcome.

AI model identity

Suggested format: identity:<provider>-<model-name>-<version>

Examples:

• identity:anthropic-claude-sonnet-4-6
• identity:openai-gpt-5
• identity:meta-llama-3-70b-instruct
• identity:google-gemini-2-ultra
• identity:custom-<org>-<model> for fine-tuned / custom models

Skill taxonomy (evolving)

• skill:code-generation-v<N>
• skill:code-review-v<N>
• skill:literature-review-v<N>
• skill:translation-<lang-from>-to-<lang-to>
• skill:legal-research-<jurisdiction>
• skill:medical-triage-<specialty>

Audit types

• audit:safety-<period> — general AI safety
• audit:alignment-<period> — alignment review
• audit:compliance-<standard>-<period> — e.g., audit:compliance-iso-27001-2026q1

See also

• Concepts: Attestations and trust — why this works
• Guide: Issue an attestation — CLI walkthrough
• Contributing — propose additions