Back to the catalog

Memory groups

OKF memory bundle for the 000_sdlc_v1 workspace - the connecting tissue across projects. Progressive-disclosure entry; open group indexes be

Open source Repository Open in the app JSON README (API)

About

# Memory groups

* [Memories](memories/index.md) - extracted, descriptive facts; one concept file per memory.
* [Assets](assets/index.md) - index concepts for cross-cutting content (transcripts, generated images + their JSON, research docs).
* [Archive](archive/index.md) - retired memories under the never-delete invariant; superseded, not erased.

# History

* [Update log](log.md) - chronological bundle history, including proof-gate verdicts.

# Rules of this bundle

* Every concept: OKF frontmatter, `type` required; store extracted facts, never raw dialogue; redact PII and secrets before persisting; carry a `source:` field naming where the fact came from (provenance).
* Retrieval: start at group indexes; grep `tags` for cross-cutting lookups; links are the edge graph.
* Curation: UPDATE on refinement, CREATE on new fact, ARCHIVE on invalidation (never delete); log every change in [log.md](log.md).

Details

Kind
OKF bundles
Topic
AI, RAG & memory
Publisher
war851
Origin
okf_github
Category
dados
Version
0.1
Forks
1
Last push
2026-07-02T19:32:22Z
Repository state
ativo
Language
HTML
License
NOASSERTION
Added
2026-09-08 22:07:22
Updated
2026-09-08 22:07:22
Origin id
war851/SDLC:wiki/index.md

README

---
type: Readme
title: SDLC Migration
description: Reusable TypeScript-native multi-project workspace on the SDLC + OKF standard, with LLM Wiki 2.0 memory and a test gate for everything that enters.
tags: [workspace, sdlc, okf, readme, migration]
timestamp: 2026-07-02T00:00:00Z
---

# SDLC Migration

A fresh, TypeScript-native, multi-project workspace built on the **SDLC + OKF + LLM Wiki 2.0**
standards, with a **Podman two-pod test gate** that proves any artifact - skill, tool, memory,
harness, model switch - minimal, well-formed, efficient, and model-agnostic (`pass^5` across
LLM families) **before** it is admitted. Use it to revamp a workspace onto the new standards,
or to test before switching LLM, harness, or SDK.

![SDLC Migration: the standard's six domain cartridges, the OKF format, the LLM Wiki 2.0 memory, the two test pods, and the prebuilt entry-gate tests](docs/assets/sdlc-migration-hero.png)

## Author's note

Google published the new SDLC and its agent whitepapers. Google Cloud published OKF, the
knowledge format that maps to Karpathy's "LLM Wiki" idea (this workspace's OKF-native
implementation is what I call LLM Wiki 2.0). The skills standard spread across vendors at the
same time, and the published examples were blunt about harness size: Anthropic reported a
150k-to-2k token reduction with skills, Vercel deleted 80% of its agent's tools.

These standards address more than size. They target the concerns that decide whether you own
what you build: switching the LLM without losing quality, staying provider-agnostic instead of
married to one vendor, and keeping your knowledge in an open, plain-text format that any tool
and any model can read. Everything you invest into a workspace, the memories, the skills, the
processes, stays yours and stays portable.

I had spent over a year perfecting my VS Code setup, my memory system, the whole thing. Now
the standards, and even my API providers, were telling me to drop most of it.

So I tested it in a separate workspace first: 45% reduction in memory size and a 96% match
rate across models, including different providers. That convinced me. Instead of patching my
old setup, I built this workspace clean on the new standards, with an entry gate, so that
everything I had, and everything future, migrates in the right way: classified, tested, proven
model-agnostic, converted to the standard, then admitted.

I am probably not the only one facing this. So here it is.

## Origins and attribution

I wrote the two PRDs, assembled published pieces (Google's SDLC and agents whitepaper corpus,
the Open Knowledge Format, Matt Pocock's skill chain), and use the result myself. The original
work here is the assembly: the distilled standard, the workspace substrate, and the entry gate.
Naming note: `SDLC` is the publication repo name and "SDLC Migration" the project title;
`000_sdlc_v1` (in the package name, container names, and internal docs) is the stable internal
workspace ID on the author's multi-workspace machine - same thing, different contexts.
Sources and attribution: [_foundation/PROVENANCE.md](_foundation/PROVENANCE.md); licensing and
third-party terms: [LICENSE](LICENSE).

## Orientation

| Where | What |
|---|---|
| [AGENTS.md](AGENTS.md) | The always-loaded core: identity, standing rules R1-R5, map. Read first. |
| [wiki/](wiki/index.md) | The memory bundle (LLM Wiki 2.0): memories, cross-cutting asset index, archive. |
| [sessions/](sessions/index.md) | Session records: log, chain-of-thought, handover per session. |
| [projects/](projects/index.md) | Registry of projects hosted in this workspace. |
| [docs/](docs/index.md) | Workspace docs: documentation method, agent seams, conflict log. |
| [tools/](tools/) | Deterministic TypeScript scripts (`npm run start-session` etc.). |
| [_foundation/](_foundation/README.md) | The original build handover bundle, kept verbatim: the standard (authority), OKF spec, skills, corpus. Historical instructions, not usage docs. |

## Prerequisites

- **Node.js ≥ 24** and git
- **Podman** (rootless is fine) with a running machine - only needed for the gate's live path
- **API keys** for the gate's proof gates: copy [.env.example](.env.example) to `.env` and fill
  in `CASE_ANTHROPIC_API_KEY` and `GEMINI_API_KEY`. Everything except gates C/D and trajectories
  works without keys (`--offline`).

## Quickstart

```
npm install                            # workspace-local toolchain, pinned
npm run typecheck && npm run selftest  # PRD1 smoke suite (sessions, memory, git step)
npm run gate -- selftest --offline     # gate pipeline, keyless and podman-less
npm run gate -- up                     # build + stand the two pods (idempotent)
npm run gate:selftest                  # full live suite: gates A-D cross-family (costs cents)
```

First time here? Follow the [first gate run tutorial](docs/tutorials/first-gate-run.md).

## Usage

```
npm run start-session   # list prior sessions (never reads their bodies)
npm run log-session     # scaffold the session record for the closing session
npm run gate -- run <artifact-dir>     # put an artifact through the entry gate
```

Agent-side, the session flows are the `/starting-session` and `/logging-session` skills. The
gate's CLI reference is generated at [docs/reference/gate.md](docs/reference/gate.md); how-tos:
[admit an artifact](docs/how-to/admit-an-artifact.md), [add a taxonomy class](docs/how-to/add-a-taxonomy-class.md).

## Standard and provenance

The design authority is the standard shipped verbatim at
[_foundation/standard/index.md](_foundation/standard/index.md). Sources and attribution for all
vendored material: [_foundation/PROVENANCE.md](_foundation/PROVENANCE.md).

More