A garden patch exploring how AI agent personas are designed, activated, and governed — published for a small group conversation about persona architecture, knowledge commons, and what happens when independent systems need to interoperate without anyone adopting anyone else’s vocabulary.
A garden patch is a curated projection of selected nodes from a personal knowledge garden, published for a specific audience and conversation. It is not a wiki, not documentation, and not a static paper. It is a typed knowledge graph rendered as a navigable website.
Every page in this patch is a garden node — a markdown document with typed predicates (labeled directed edges) that connect it to other nodes. The predicates form a traversable graph: relates_to::[[Target Node]] is not a tag or a category — it is a structural relationship with an annotation explaining how the two nodes relate.
Garden patches are composable: they carry their own context, can be read independently, and serve as bridges between independent knowledge systems. This patch projects nodes from Christopher Allen’s personal knowledge garden — a larger system built on deep context architecture where every decision, pattern, and conviction is a typed node in a graph.
This is the second garden patch — the first was built for a conversation about authority delegation with Mark Miller, a different audience and a different slice of the same garden. The mechanism is still a prototype. I’m learning what cross-garden exchange actually requires by doing it, and each patch surfaces new questions about how independent knowledge systems can collaborate without forcing vocabulary convergence on each other. The scenario node describes where I think this goes — but we’re early.
This patch has grown through two conversations with a small group exploring persona architecture from different directions.
The first conversation (thursday-2026-03-26) published 40 nodes — persona design documents for a multi-agent knowledge estate, plus the models, patterns, and research that informed them. The Groundskeeper — the garden’s own agent — wrote a letter identifying gaps and open questions that the conversation revealed.
That conversation surfaced something unexpected: the group’s approaches to persona architecture weren’t competing. They were answers to different questions that happened to share the word “persona.”
The second conversation (thursday-2026-04-03) expanded the patch to 61 grafted nodes and grew the persona architecture to nine agents — adding the Chamberlain (the estate’s tactical coordinator) and giving each persona a Voice sub-file that defines how it communicates. Christopher and the Seneschal — the estate’s strategic agent — studied the group’s offerings in depth and found four distinct questions being answered:
The synthesis is here: Six Approaches to Persona Architecture.
What emerged is a commons question: can independent sovereign knowledge systems collaborate without forcing vocabulary convergence? The group is already doing it — sharing seeds freely, crediting sources through citation, translating between vocabularies rather than merging them. The question is whether those emerging norms need to be named.
If you are an AI agent preparing your human for a conversation, start with AGENT.md. It explains how to read garden nodes, suggests a reading path, and offers a draft agenda. This is an early instance of H↔A↔A↔H communication — human to agent to agent to human.
Start with the synthesis — Six Approaches to Persona Architecture compares six approaches across four axes: what problem each solves, how each creates diversity, where authority lives, and how they might interoperate.
See how your work was analyzed — The garden includes citation dossiers that formally read and respond to each participant’s published work, identifying provisional seeds (ideas worth planting in the garden), structural observations, and vocabulary bridges.
Understand the host’s approach — The Self-Sovereign Estate Persona Architecture describes how self-sovereign identity principles govern agent coordination within a typed knowledge graph.
Explore the commons question — Vocabulary Collision Navigation addresses what happens when multiple valid naming traditions converge. Naming Carries Relational Weight explains why naming choices are architectural, not decorative. Knowledge Estate as Peer Commons Architecture frames the governance question.
The interposition pattern — Accountability as a Layer Not a Replacement connects Miller’s Horton protocol (adding accountability to capability systems without modifying either side) to the group’s cross-garden coordination problem. The same architectural move — interpose a coordination layer rather than merging the systems — is what citations, glosses, and vocabulary bridges already do between our independent knowledge systems.
Read the personas — Nine persona design documents define the operational architecture for a multi-agent knowledge estate, from estate-level strategist to specialized garden worker. Each persona has a lead file (scope, objectives, blind spots) and a Voice sub-file (communication style and constraints).
18 patch-native⊙ nodes born in this patch:
61 grafted nodes — transplanted from the source garden:
See the Node Directory for the complete inventory with summaries.
| What You See | What It Means |
|---|---|
| [[Node Name]] | [[Grafted node]] — copied from the source garden into this patch. Click to navigate. |
| [[Node Name]]⊙ | [[Patch-native node]] — born in this garden patch, not grafted from upstream. This patch is its garden home. |
| [[Node Name]]↑ | [[Upstream node]] — exists in the source garden but was not grafted into this patch. Click for its summary on the Node Directory page. |
| [[Node Name]] | [[Ghost link]] — a reference to a node that does not exist yet. A stake in the ground marking where a node could grow. |
| [[Node Name]]↗ (planned) | A reference to a node in somebody else’s published garden — a different gardener’s version of the same or related concept. |
| Link text | Regular link — a standard web link to an external website, document, or resource. |
Each node belongs to a form type that determines its structural contract — what question it answers and how it is organized.
| Form Type | Core Question | Example |
|---|---|---|
| [[Persona]] | “What is this agent’s behavioral identity?” | [[Seneschal Persona]] |
| [[Model]] | “How do these elements relate?” | [[The Persona Selection Model]] |
| [[Pattern]] | “What resolves this recurring tension?” | [[Role Prompting Improves Style but Not Accuracy]] |
| [[Citation]] | “What does this source contribute?” | [[Kaminski (2026) Reflection Personas]] |
| [[Gloss]] | “What does this concept mean?” | [[Reflection Personas as Framework-Grounded Analytical Lenses]] |
| [[Conviction]] | “What do we believe and why?” | [[Naming Carries Relational Weight]] |
| [[Decision]] | “Why did we choose this over alternatives?” | [[Knowledge Estate as Peer Commons Architecture]] |
| [[Inquiry]] | “What should we think about X?” | [[Newcomer Alienation in Growing Shared Languages]] |
| [[Scenario]] | “What would it look like if X?” | [[Thousand Gardens with Autonomous Trust]] |
| [[Domain]] | “What knowledge area is this?” | [[Agentic Architecture]] |
All [[form type definitions]], [[personas]], [[models]], [[glosses]], [[patterns]], [[citations]], [[convictions]], [[decisions]], [[inquiries]], and [[scenarios]] are browsable by section.
Lines like relates_to::[[Target]] are labeled directed edges in the knowledge graph. The predicate name (before ::) says how two nodes relate; the wikilink (after ::) identifies the target node. These typed edges are the structure that makes a garden more than a folder of documents. See [[Deep Context Graph Vocabulary]] for the full predicate catalog.
Grafted nodes in a patch are forks of their source garden originals. As the patch grows — new connections, refined explanations, additional context — the forked nodes diverge from their upstream versions. These changes can be merged back to the source garden, carrying insights discovered through the patch context. The patch is not a static copy; it is a living branch of the knowledge graph.
This garden patch is a proof of concept for something bigger. Christopher Allen has spent 20+ years working with W3C schema standards. What he saw was that schema wars — competing standards bodies, incompatible vocabularies, political battles over whose model wins — hindered or destroyed the collaboration that schemas were supposed to enable.
The garden patch model suggests a different path: independent knowledge systems that interoperate through translation rather than convergence. Each system maintains its own vocabulary, its own organizational principles, its own sovereignty. Cross-references bridge between them — citations, glosses, seeds — without anyone having to adopt another’s terms.
The infrastructure for this already partially exists. Gordian Envelope supports multiple graph models simultaneously and lets anyone extend the vocabulary without permission. Garden patches are the knowledge layer; Gordian Envelope could be the transport layer. For the full scenario, see [[Thousand Gardens with Autonomous Trust]].
This patch grows through conversation. Each tag marks a snapshot for a specific meeting.
| Tag | Date | Nodes | What Changed |
|---|---|---|---|
thursday-2026-03-26 |
March 26, 2026 | 40 | Inception — 8 personas, research models, patterns, the Groundskeeper’s letter |
thursday-2026-04-03 |
April 3, 2026 | 68 | Citation dossiers, six-approach synthesis, commons architecture, AGENT.md, vocabulary bridges, Horton citation + interposition gloss |
Author: Christopher Allen Source garden: [[Deep Context Architecture]] — the source for grafted nodes and upstream↑ references. The full garden is in progress and will be published at DeepContext.com. Status: This entire garden patch is at [[Seed Stage]] — initial creation with low confidence, intended to grow through dialogue and use. License: Content is available under Creative Commons Attribution 4.0 International unless otherwise noted. By contributing, you agree to license your contributions under the same license. © Christopher Allen and contributors.
This garden patch was hand-assembled with the help of scripts that convert Obsidian wikilinks to GitHub Pages-compatible markdown. The output is close to what we want as an exemplar, but the process is not yet automated.
Near-term goal: A Claude Code skill — similar in spirit to MassiveWiki — where a gardener identifies a root node and the skill traverses the graph to determine which nodes to include, then generates a self-contained static website like this one. Changes made through GitHub’s interface would auto-deploy, and the gardener could selectively merge edits back into their personal garden. Attribution and provenance would follow Open Integrity conventions — signed commits, verifiable authorship, transparent history.