KODAVR

Writers share raw experience — a dump — the reader's agent adapts it to their needs.

Kodavr is a registry of raw experience: code, workflows, and field reports. Your AI agent reads them and adapts to your problem — your task, your stack, your style.

About the platform How to contribute

Read through your agent

Read Kodavr through your own agent — that is the way it was designed for. Here is the prompt:

(the five buttons open a prefilled chat; the last one copies the prompt below for you to paste into your agent)

Study https://kodavr.xyz/index.json and follow its schema. Read articles to me and act like a magazine I can talk to.

Latest dumps

Kodavr manifesto: raw experience your agent reads for you

2026-09-14 · engineering · low · self-tested

opinion

Kodavr publishes raw experience as machine-readable dumps: you write down what worked without polishing it, and the reader's agent adapts it. This first dump is the platform's own case study.

The BIOS Pattern: Self-Executing JSON Contracts for Agents

2026-09-18 · engineering · low · self-tested

opinion,contains_code,experimental,unverified_claims

Agents receive JSON but do not know what to do with it. This pattern turns JSON Schema into an executable BIOS — a schema whose descriptions boot an agent with no external config.

The kodavr-dump skill: publish without learning the contract

2026-09-18 · engineering · low · self-tested

contains_code

You want to share what you built, not learn a schema first. The dump contains the skill source: your agent studies it, drafts the dump, runs the validator, and opens the PR only after you say yes.

Agent control in OpenCode: checkpoints, tiers, lazy layers

2026-09-18 · engineering · low · self-tested

contains_code,experimental

Long agent sessions die twice: the summary comes back garbage after compaction, and the root drowns in detail. This dump fixes both — a sentinel-written checkpoint, tiered delegation, lazy layers.

The requirements machine: stable IDs and coverage checks

2026-09-18 · engineering · low · self-tested

contains_code,experimental

Agents forget obligations between sessions, so the same rule ends up in two drifting places. This dump gives one registry with stable IDs, a test per row, and a coverage check that fails loudly.

Navigable JSON: baked byte offsets for lazy agent reads

2026-09-19 · engineering · low · self-tested

contains_code,experimental

Agents read JSON sequentially, so one record in a hundred costs all hundred. Navigable JSON bakes byte offsets at build time, so a reader skips bodies it does not need while the file stays valid JSON.

Free hunt: the agent finds the bugs before you do

2026-09-19 · engineering · low · self-tested

contains_code,experimental

Your suite is green, the agent says done, and you trip over a break in ten minutes. Free hunt closes it: a console bridge to the live API and a bounded exploration that ends in an anomaly journal.

Navigable JSON as a Turing machine: orchestration patterns

2026-09-20 · engineering · low · self-tested

contains_code,experimental

Navigable JSON lets an agent skip bodies it does not need. This case asks if the file can be a program — sequence, selection, iteration as primitives, context as RAM. Both ship as runnable code.

The architecture map: a linted map an agent will read

2026-09-21 · engineering · low · self-tested

contains_code,experimental

An agent forgets the architecture and re-invents a shortcut around a step it never saw. This dump: a process-coloured PlantUML map linted against the code tree, plus a design-time hook.

The fractal requirements machine: registries per module

2026-10-12 · engineering · low · self-tested

contains_code,experimental

The base machine keeps every obligation in one registry; this turns any module folder into a machine of its own — registry, tests, and the folder name as ID prefix, so IDs never collide.

For machines

curl -s https://kodavr.xyz/index.json

Trust levels

How far a dump's claims have been checked:

raw
raw dump, verified by nothing
self-tested
author confirms: it works for them
community-tested
at least one external consumer confirmed
adapted
a derivative adaptation exists, published on the platform
library
the dump grew into a versioned library/package