Web7 RFCs

How the RFC process works

RFC stands for Request for Comments. It is a formal proposal to add, change, or remove a normative behaviour in the Web7 protocol. Any developer can propose one.

Active & Draft RFCs

Recent Updates

Significant changes to the RFC register and the protocol reference implementation.

Community Questions

Common questions from developers building on Web7. Answers reflect current Draft RFC decisions and may change before Final status.

Agent Mesh Protocol — AMP v0.1

Motivation

Web3 moves tokens. Web7 moves intent. No existing protocol expresses outcome-based delegation between autonomous agents with reputation, policy, and verifiable inference. AMP fills this gap.

What AMP defines (normative)

• Message envelope schema — JSON with DID signing and canonical serialisation
• 7 protocol verbs: discover · negotiate · delegate · execute · settle · attest · revoke
• 4 settlement rails: amp-escrow · amp-outcome · amp-subscription · amp-split
• Attestation format (W3C VC compatible + ZK-ML proof optional)
• Reputation update formula: outcome_value × principal_stake × domain_match × age_decay
• Slashing triggers (6 conditions) and distribution rules

Open questions (your feedback needed)

Kynetra Prime Spec — v0.1

What Prime defines (normative)

• Agent object schema: id · principal · skills · wallet · memory · policy · reputation
• Agent lifecycle: spawn → authorise → route(task) → delegate → settle → record → learn → retire
• Trust model: zero implicit trust — every action signed, every route policy-checked, every outcome logged
• Skill registry: discovery by tag / semantic match / reputation floor
• Memory governance: what context an agent may access per intent scope
• Policy engine: org rules, compliance, jurisdiction filters

Proof-of-Outcome (PoO) Standard

What PoO defines (normative)

• The 8-check verification predicate: principalSig · delegationSig · agentSig · zkml · policy · stake · sla · attestors
• Check weights (total 100): principalSig(25) · delegationSig(20) · agentSig(15) · zkml(15) · policy(10) · stake(5) · sla(5) · attestors(5)
• Passing threshold: weighted score ≥ 80 with all required checks passing
• ZK-ML stub format for v0.1 (k-of-n attestors); full ZK circuit spec post-L0
• Slashing triggers (6 conditions) and distribution rules (principal vs treasury allocation)
• Settlement value formula: score/100 × maxAmount for partial outcomes

Open questions

Accountable Intelligence Graph (AIG) Specification

What AIG defines (normative)

• Node types: intent · delegation · inference · outcome — each carries a signed AMP envelope
• Edge semantics: causation (A caused B) · delegation (A delegated to B) · attestation (A attests B)
• Merkle commitment: FNV-1a leaf hashing, deterministic node ordering, root commitment format for L0
• Lineage query API: lineage(outcomeId) returns full causal chain to root intent
• Audit export: signed subgraph as verifiable JSON-LD document
• Tamper evidence: any modification to a committed node is detectable via Merkle proof mismatch

Open questions

Self-Evolution Protocol (SEP) v0.1

Motivation

Every Web7 deployment accumulates a rich AIG — a signed, tamper-evident record of every intent, delegation, inference, and outcome. Today this graph is queried for audit and compliance. SEP specifies how the same graph is queried for improvement: identifying what the system does well, what it fails at, and what modules it needs that don't exist yet. The result is a framework that gets smarter with every deployment, without external supervision.

The five SEP components (normative)

ImprovementProposal schema (normative)

{
  id:         string,             // "prop-{ts}-{nonce}"
  type:       "prompt" | "workflow" | "module" | "conformance",
  agentDid:   string,             // did:w7 of the agent being improved
  evidence:   string[],           // outcomeId[] that triggered this proposal
  suggestion: string,             // human-readable improvement text
  confidence: number,             // 0–1, capped at 1.0
  meta:       Record<string, unknown>,
  createdAt:  ISO8601
}

TestVector schema (normative)

{
  id:          string,            // "auto-{check}-{ts}"
  name:        string,            // "auto.{check}.failure"
  description: string,
  check:       string,            // failed PoO check name
  evidence:    PoOResult[],       // raw results that surfaced this vector
  emittedAt:   ISO8601
}

PatternMiner AIG query (normative)

PatternMiner operates on the output of aig.walk() or aig.lineage(rootId). It counts occurrences of each intent.action value across all intent nodes in the result set. Patterns with count ≥ minFrequency (default: 3) are surfaced. Patterns with successRate ≥ 0.80 and count ≥ 5 are classified as high-value module candidates.

ProtocolAdaptor binding rules (normative)

• Each OpenAPI operation maps 1:1 to one AMP skill (skillId = agentDid + "#" + operationId)
• The adaptor handler wraps the HTTP call and returns { ok: boolean, status: number, result: unknown }
• All adaptor skills must be registered with Prime's skill registry via registerSkill() before use
• PoO conformance applies to adaptor skills — outcomes must be wrapped with createOutcome() and verified with verifyPoO()

Open questions (your feedback needed)

did:w7 Method Specification

What did:w7 defines (normative)

• DID syntax: did:w7:{slug} where slug is an unreserved string
• Entity namespaces: human (did:w7:alice) · Prime (did:w7:prime/main) · skill (did:w7:skill/name) · model (did:w7:model/name@sha256:hash) · org · attestor
• CRUD operations: create · resolve · update (key rotation) · deactivate — with signature requirements
• Key types: Ed25519VerificationKey2020 (required) · JsonWebKey2020 (optional)
• W3C extensions: w7.reputation · w7.stake · w7.skills in DID document
• L0 anchoring: in-memory registry (v0.1) → L0 DAG anchored registry (production)

Open questions

Propose an RFC

Have an improvement, extension, or new primitive? The RFC process is open to everyone.

git clone https://github.com/HyperBridge/w7f-core
cd w7f-core && git checkout -b rfc/my-proposal

cp rfcs/0000-template.md rfcs/0003-my-proposal.md
# Fill in: title, motivation, specification, security, open questions