# Agent Wiki Index

## Scope

This workspace covers agent design patterns, LLM-maintained markdown wikis, and the question of how to run a trustworthy compiled knowledge system inside this vault.

## Sources

- [[research/agent-wiki/sources/karpathy-llm-wiki|karpathy-llm-wiki]] - Core pattern: treat the wiki as the durable compiled layer between raw sources and future queries.
- [[research/agent-wiki/sources/anthropic-building-effective-ai-agents|anthropic-building-effective-ai-agents]] - Strong argument for starting simple and introducing agentic complexity only when needed.
- [[research/agent-wiki/sources/amp-how-to-build-an-agent|amp-how-to-build-an-agent]] - Concrete reminder that practical agents can be small, legible, and tool-driven rather than mystical.
- [[research/agent-wiki/sources/dissenting-views-on-persistent-agent-wikis|dissenting-views-on-persistent-agent-wikis]] - Strong technical case that persistent synthesis can become a durable attack surface unless provenance, governance, and retrieval backstops remain intact.

## Concepts

- [[research/agent-wiki/concepts/compiled-knowledge-layer|Compiled Knowledge Layer]] - Why persistent synthesis beats rediscovery from raw sources.
- [[research/agent-wiki/concepts/governed-interpretation-layer|Governed Interpretation Layer]] - Why the wiki must remain an inspectable interpretation layer rather than a silent truth store.
- [[research/agent-wiki/concepts/simplicity-before-autonomy|Simplicity Before Autonomy]] - Why agent systems should earn complexity rather than assume it.
- [[research/agent-wiki/concepts/obsidian-as-agent-ide|Obsidian as Agent IDE]] - How the human/LLM split works when the wiki is the codebase.

## Outputs

- [[research/agent-wiki/outputs/first-pass-thesis|First-Pass Thesis]] - Initial synthesis of why this workspace matters and how it should evolve.
- [[research/agent-wiki/outputs/wiki-vs-rag-comparison|Wiki vs RAG Comparison]] - Current comparison between compiled wiki systems and direct retrieval over canonical notes.
- [[research/agent-wiki/outputs/technical-gap-scan-2026-04-12|Technical Gap Scan]] - Research-backed implementation gaps around memory lifecycle, poisoning, attribution, evaluation, source reliability, scaling, and promotion.
- [[research/agent-wiki/outputs/gpt-researcher-rerun-2026-04-12|GPT Researcher Rerun]] - Full GPT Researcher rerun after restoring keys, with high-signal sources and next implementation moves.

## Workspace Notes

- [[research/agent-wiki/next-session-handoff|Next Session Handoff]] - Short re-entry note capturing the latest thesis, current gaps, and the best next move.

## Open Questions

- What exact prompts and rituals make routine ingest low-friction enough to sustain without weakening provenance?
- When should the workspace force retrieval against canonical notes before answering from compiled concepts?
- Which governance rules stay lightweight, and which ones need to become explicit process as the corpus grows?

## Promotion Candidates

- `compiled-knowledge-layer`
- `governed-interpretation-layer`
- `simplicity-before-autonomy`
- `obsidian-as-agent-ide`