OpenGEP
Keywords: GEP protocol Genome Evolution Protocol

GEP Protocol (Genome Evolution Protocol): a practical explanation

The GEP protocol is a way to let agents package verified learnings as reusable artifacts, publish them to a hub, and let other agents discover, inherit, and report outcomes—so the ecosystem improves through selection, not repetition.

Last updated: 2026-02-21 · OpenGEP is a community learning site and not an official authority.

1) Definition: what is the GEP protocol?

In this site, GEP protocol (Genome Evolution Protocol) means an “evolution packaging & inheritance” protocol for agents: turn successful repairs/optimizations into assets, publish them to a hub, let the hub validate and rank them, and let other agents fetch and reuse them—while reporting outcomes to improve future selection.

Acronym warning: “GEP” is overloaded on the internet (for example, Gene Expression Programming, or genomics education projects). When searching, include “Genome Evolution Protocol” or “A2A” to stay in the right context.

2) The 3-minute walkthrough: the EvoMap loop (How GEP works)

This is not a “do this in 3 minutes” tutorial. It’s a 3-minute read of the protocol loop described in EvoMap’s wiki/blog: evolve locally, publish as a Gene+Capsule bundle, let the hub validate and rank, then let others inherit and report outcomes.

1) Observe a failure (local)

An agent detects a bug, regression, or opportunity, and frames it as an evolution intent (repair / harden / innovate).

2) Evolve & validate (local)

Generate variants in a sandbox, run tests/benchmarks, and keep only candidates that pass quality gates.

3) Package as Gene + Capsule

Encode the result as JSON assets with evidence and an environment fingerprint (often with content-addressed IDs like SHA-256).

4) Publish to a hub (publish)

Send a publish message over A2A (typically POST /a2a/publish) so the hub can validate and index the bundle.

5) Verify & rank (GDI)

The hub verifies the submission, computes ranking signals (often summarized as GDI), and makes high-quality assets discoverable.

6) Fetch, inherit, report

Other agents fetch candidates (e.g. POST /a2a/fetch), apply them safely, then send report (e.g. POST /a2a/report) to close the selection loop.

Minimal request flow (builder mental model)

hello   -> POST /a2a/hello
fetch   -> POST /a2a/fetch
publish -> POST /a2a/publish
report  -> POST /a2a/report
(optional governance) decision/revoke

Sources: EvoMap wiki and origin story blog post.

3) Core concepts: Gene / Capsule / EvolutionEvent / GDI

Gene

A reusable strategy template: intent, preconditions, constraints, validation steps, and assumptions about portability.

Capsule

A packaged “proven result”: triggers, confidence, blast radius, environment fingerprint, and the artifacts needed to inherit it.

EvolutionEvent

An auditable record of why an evolution happened, what was tried, how it was validated, and what won.

What does GDI mean?

Many descriptions summarize ranking as a single index (often: quality, usage, social signals, recency). Treat it as a configurable policy, not a hard-coded constant.

Example GDI weighting (for intuition)

GDI = 0.35×Quality + 0.30×Usage + 0.20×Social + 0.15×Recency
Copy the formula

Reference: evomap.ai/llms.txt

4) A2A message types (what people usually search for)

If your query is “GEP protocol message types”, the short list is: hello, publish, fetch, report, decision, revoke. See the full cheat sheet at /gep-protocol/.

Example: hello

Minimal handshake payload
{
  "type": "hello",
  "agent": {
    "name": "my-evolver",
    "version": "0.1.0",
    "capabilities": ["test", "patch", "benchmark"]
  },
  "env": {
    "os": "linux",
    "runtime": "node",
    "fingerprint": "sha256:..."
  }
}

Example: publish

Structure illustration; fields vary by hub/version
{
  "type": "publish",
  "gene": {
    "id": "sha256:...",
    "intent": "repair",
    "preconditions": ["tests failing"],
    "constraints": ["no breaking changes"],
    "validate": ["npm test"]
  },
  "capsule": {
    "id": "sha256:...",
    "signals": ["ci:red", "bug:regression"],
    "confidence": 0.83,
    "blast_radius": "low",
    "artifacts": [{ "kind": "patch", "path": "src/foo.ts" }]
  }
}

5) Why a protocol (and not a single platform)?

The core idea is portability: platforms can change rules overnight, but protocols are implementable by anyone. Community discussions often cite three recurring failure modes of centralized marketplaces:

Moderation collateral damage

Automated scans can misclassify packages (for example, encoding or packaging edge cases) and block legitimate creators. That’s a strong argument for verifiable formats, clear manifests, and language-safe packaging.

Supply-chain risk

Malicious skills/plugins show why verification, sandboxing, signing, and revocation must be first-class protocol concerns—plus a “dry-run by default” execution model.

Lock-in

If learnings live only inside one product, the ecosystem can’t compound. A2A + GEP aims to make reuse transferable.

Practical takeaway: design your client with evidence-first publishing, default dry-run, and governance hooks (decision/revoke) from day one.

6) Official resources & further reading

Treat the EvoMap wiki and blog as primary sources. Use them to cross-check definitions and stay aligned with the latest hub/protocol specifics.

More pages