Context Engineering in an LLM Harness
The article introduces a context engineering approach for LLM harnesses that defers large domain knowledge to on‑demand indexes rather than embedding it in the system prompt. It describes a typed ontology graph with domains, entities, and relationships as the first layer of this architecture. The method aims to reduce token costs, improve model accuracy, and keep information reachable via stable addresses.
- ▪Embedding all domain knowledge in the system prompt becomes costly as conversations grow, prompting the need for a deferral strategy.
- ▪The proposed harness maintains five small indexes—ontology, tool discovery, values by reference, memory, and skills—each pulling payloads only when required.
- ▪The ontology is a typed graph comprising 15 domains, 28 entities, and 12 relationships, providing structured context for tasks like issue‑tracker queries.
- ▪Deferring context reduces token usage and attention competition, leading to higher accuracy while preserving full information accessibility.
- ▪The design relies on validated references and stable addresses to ensure that shrinking the standing context removes noise without discarding needed data.
Hacker News (AI / LLM) files mainly under ai. We currently carry 3,828 of its stories.
Story provenance
Source · retrieval · rights · ranking — open for full record
inspect →
Story provenance
Attribution is not the same as permission. This drawer separates discovery metadata, excerpts, WeSearch-generated summaries, reuse status, and whether the publisher receives the visit. Nothing here claims a legal grant the publisher has not made.
Record
| Original publisher | Udnes |
| Canonical URL | https://udnes.dev/posts/context-engineering-harness-part-1-ontology/ |
| Publication time | Thu, 06 Aug 2026 07:25:19 +0000 |
| Retrieval time | 2026-08-06T07:40:43.661Z |
| Last seen | 2026-08-06T07:40:43.661Z |
| Headline source | Publisher (no WeSearch rewrite) |
| Excerpt source | publisher body |
| Excerpt method | First ~120 words (~800 chars) of extracted publisher body, fair-use limited. |
| Summary | WeSearch · cerebras-chat (WeSearch summarizer) |
| Summary source text | contentText |
| Citation coverage | Summary is a WeSearch-generated derivative; primary citation is the original publisher URL. |
| Cluster | Y2O5XNHiOfWk · 1 stories |
| Cluster logic | Grouped by semantic title/content similarity across sources within a rolling window. Same-publisher template collisions are excluded from coverage comparison. |
| Ranking reason | Story pages are not engagement-ranked. Hub feeds use recency, with optional source-diversified chronological ordering (cap consecutive stories per source). No personalized ranking. |
| Publisher visit | Yes — open original |
| Substitutes article? | No — link-out required for full text |
Rights status (four layers)
WeSearch handling by dimension
| Indexing | May the item be indexed (stored, ranked, made findable)? | Allowed |
| Snippet | May a short excerpt of the publisher's text be shown? | Allowed |
| AI summary | May WeSearch generate its own short summary of the article? | Limited |
| Retrieval / RAG | May the content be exposed for third-party retrieval-augmented generation? | Not asserted |
| Model training | May the content be used to train AI models? | Not asserted |
| Commercial reuse | May the content be reused commercially? | Not permitted |
Basis: Derived from the published RSS/Atom feed. Contact: [email protected]. Reviewed: 2026-07-24.
Opening excerpt (first ~120 words) tap to expand
Context Engineering in an LLM Harness — Part 1: Ontology July 31, 2026 Context Engineering in an LLM Harness · 5 parts Ontology Tool Discovery Values by Reference Memory Skills Next: Tool Discovery → Imagine embedding an LLM agent in an issue tracker. A user opens the assistant inside a release project and asks: “which open issues are blocking the release?” The model needs product-specific knowledge: how projects, issues, workflows and dependencies fit together; which status names this team uses; and how internal ids map to the names people see. Echoing a raw id back at the user is still a bug, not an answer. The obvious move is to paste all of that into the system prompt. Everybody’s first harness does this, ours included.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Udnes.
Discussion
0 commentsMore from Udnes
