---
source_hash: 4407b121230f641b7fea16d3cacda741e535d39682ce38d7d3161f1a69378d0d
post: /posts/workflows/aris-space-icm-knowledge-bank
---

# Context Pack: Ari's Space ICM Knowledge Bank

Reference for the August 2026 library bundle. The supplied zip was inspected: 84 entries, 64 episode cards, built 6 August 2026. It is a dated snapshot, not an automatically refreshed copy of every current essay. The live essays remain authoritative.

## Verbatim excerpts

> The full essays stay on the site. The folder helps you find them.

> Catalog points. Cards excerpt. The URL is the book.

> The best match can also be number one in the list; say so.

## Download and form

Public ungated zip: /downloads/aris-space-icm-knowledge-bank.zip. The article retains its Download component, directory tree, answer-format example, ask→bank→site diagram and //A<3 sign-off code block.

ICM means Interpretable Context Methodology, attributed to Jake Van Clief and collaborators. This is its knowledge-bundle form: small routing files and saved Markdown help an agent walk a public teaching corpus. It is not a vector database, multi-agent framework, private operations dump, offline whole-site copy or replacement for /documents.

## Structure and answer contract

CLAUDE.md holds orientation/answer rules; CONTEXT.md maps shelves; AGENTS.md points non-Claude seats at the same instructions. _index provides log/by-topic catalogues; _meta/schema.md defines cards; _shared/advice-voice.md holds guidance; corpus/_index.md records the harvest. bundle/voice, bundle/dispositions and bundle/episodes hold orientation, task routers and cards respectively.

Each card has title, live URL on its own line, one paragraph of context, when to recommend it and related cards. Routers cover folders/loops, orchestration, memory, tools, outcomes, workspaces, dispatch and courses. Topic shelves include workflows, tools, dispatch, memory, workspaces, ARI-OS and doctrines.

The response is Best match, then Top three. Supply title, URL and a short paragraph explaining fit for each. Best match may also be number one; acknowledge that. Recommend actual cards rather than inventing essays or dumping the corpus.

## Three use paths

1. Local agent: unzip and use a host able to read files. Load CLAUDE.md, CONTEXT.md, _index/by-topic.md, a disposition and one to five cards. The example asks what to read about losing work in chat history.
2. Human: open _index/by-topic.md, select a shelf, read three cards and follow URLs.
3. Cloud ICM desk: the original August account described Jake Van Clief's Clief Notes/Eduba community platform, with uploaded folders, one strong model/harness, per-workspace containers, Markdown rendering, file edits, package installation, Python/Playwright and multi-person sharing. VIP/alpha access and covered AI cost were rollout details, not promises about present access, price or capability. The folder remains the portable starting material; it does not grant runtime permissions.

Original signup: https://docs.google.com/forms/d/e/1FAIpQLSdSnToxclxt8EgoB2GizrEvHP7gzOedVcRXdCkpi1SZwR8ZfA/viewform. Contact: info@eduba.io.

Regenerate and check the bundle when the library changes, then replace the zip. Until that happens, retain the snapshot date and use the live essay when a card's wording differs.