Literature Landscape for miaco: Prompt Decomposition, Inquiry Routing, Episodic Memory, and Indigenous-Governed Agent Workflows

Overview

This report maps the architecture of miaco—the Engineering World CLI agent in the mia-co npm package—onto existing academic and practitioner literature, focusing on prompt decomposition, inquiry routing, episodic memory, and Indigenous-governed workflow models. The goal is to identify nearby fields, representative works, and language you can cite when framing miaco as a contribution to AI agent research rather than just tooling.[^1][^2][^3]

System Context: miaco and the PDE → Inquiry → QMD → STC Spine

The mia-co repository positions miaco as a terminal agent for schema design, structural tension charting, and system architecture, operating within a three-world ecology (Engineering, Story, Ceremony). Within this ecology, the attached PDEs define a pipeline: decompose → emit inquiry → ground locally via QMD → enrich current reality → advance with directional governance using a Medicine Wheel frame. This pipeline aligns miaco with research on multi-agent orchestration, query routing, and episodic memory in conversational agents, while its Indigenous framing connects to decolonial design and Medicine Wheel-based evaluation frameworks.[^4][^2][^5][^6][^3][^7][^8][^9]

Prompt Decomposition and Inquiry Routing

Prompt Decomposition and Sub-Question Generation

Work on query transformation and prompt decomposition in RAG and agent pipelines provides a direct backdrop for miaco’s Prompt Decomposition Engine (PDE). LlamaIndex’s “Query Transform Cookbook” formalizes transformations such as routing, query rewriting, and sub-question decomposition as first-class components in a tool-based pipeline, closely paralleling PDE’s move from monolithic prompts to structured sub-intents. Research and tutorials on prompt chaining similarly show that decomposing tasks into sequential prompts improves reliability and user preference over single-shot prompts, especially in summarization and multi-step reasoning tasks.[^2][^10][^11]

Multi-Agent Routing Architectures

Multi-agent routing prompts in practice-oriented systems (e.g., Docsbot’s Multi-Agent Query Router and various open-source orchestrators) treat routing as a learned or rule-based policy that selects which specialist agent should handle a query based on context, attachments, and required data freshness. Academic lectures and practitioner guides on agent routing describe patterns like mixtures of experts, self-reflection, and tool selection as key to improving multi-task performance. These works validate miaco’s Event-driven Inquiry Routing PDE, which casts PDE as an event emitter whose ambiguities, assumptions, contradictions, and low-confidence segments are turned into typed inquiry events with defined payload schemas and subscriber contracts.[^12][^13][^11][^1]

Structural Tension and Governance as Routing Policy

Miaco’s use of Structural Tension Charts (STC) and Medicine Wheel directions to govern when and how inquiries progress from East (vision) to South (analysis) adds a governance layer largely absent from mainstream routing literature. Contemporary agent orchestration examples (e.g., Maestro or multi-agent coding tutorials) introduce step-gated workflows and hard checkpoints but frame them in purely engineering terms. Miaco’s directional governance PDE instead encodes progression conditions as stage gates and hold states grounded in Indigenous directional teachings, situating routing and progression within a ceremonial logic rather than a solely performance-driven pipeline.[^5][^7][^13][^8][^14][^4]

Local-First Retrieval and QMD-Grounded Inquiry Resolution

Local-First Retrieval and Query Heuristics

Miaco’s QMD-grounded Local Inquiry Resolution PDE defines QMD as the first local grounding layer for routed inquiries, specifying inquiry classes, collection-routing policies, and numeric escalation thresholds. This corresponds to local-first RAG patterns where systems query local indices (project documents, workspace, prior sessions) before escalating to global search or external APIs, often tuned via class-specific heuristics and score thresholds. The PDE’s emphasis on explicit heuristics, collection metadata, and validation via recall/precision over labeled query sets mirrors best practices in retrieval system design, including measuring precision/recall and constructing small ground-truth corpora for policy tuning.[^15][^10]

Multi-Engine Terminal Orchestration

Miaco’s architecture explicitly coordinates multiple LLM engines—Gemini, Claude, and Copilot—via terminal commands and headless clients, raising questions of cost/quality routing and parallel vs handoff scheduling. Practitioner work on orchestrating multiple AI coding agents shows similar patterns: specialized agents for frontend, backend, and testing, parallel execution of sub-tasks, and human-in-the-loop supervision to manage failures. These systems reinforce miaco’s design choice to treat engine selection and query routing as governed by explicit policies rather than monolithic “best model per query” heuristics, aligning the codebase with ongoing work on multi-provider, multi-agent orchestration.[^3][^16][^13][^14][^2][^12]

Episodic Memory and Polyphonic Artifact Parsing

Episodic Memory Models for Conversational Agents

Episodic memory architectures for conversational agents aim to preserve a sense of autobiographical continuity, storing structured “episodes” that can be retrieved and replayed to maintain context across interactions. Early work on episodic memory for Embodied Conversational Agents (ECAs) highlights the importance of structured episode representations and retrieval mechanisms that go beyond simple logs or embeddings. More recent proposals, such as bundled episodic memory architectures, encode meaningful conversational events as structured memory bundles at write time, preserving artifact identity, provenance, semantic representation, and collaboration metadata, and retrieving entire bundles based on new inputs.[^17][^18]

Miaco’s Polyphonic Artifact Parsing and Episodic Kinship Memory PDE mirrors these approaches by designing how talking-circle or polyphonic markdown becomes episodic kinship memory, including memory schemas, parser rules, and PDE/session linkage models. The action stack—gather examples, define memory objects, specify parser rules, map linkage to PDE/session/orchestration—aligns with episodic memory work that emphasizes example-based schema design, parsing pipelines, and explicit linkage to dialog or session IDs.[^19][^18][^2][^17]

Polyphony, Talking Circles, and Relational Memory

The focus on polyphonic artifacts and talking-circle transcripts extends episodic memory beyond individual user-bot dialog into multi-voiced, relational contexts. While mainstream episodic memory literature centers on single-user interactions, work on character embodiment and conversational AI for rich, persistent characters increasingly considers multi-party and narrative contexts where memory must preserve relationships among multiple speakers and artifacts. Miaco’s notion of “episodic kinship memory” explicitly encodes actors, roles, and kinship metadata, overlapping with research that treats conversational memory as socially embedded rather than purely transactional.[^18][^20][^19][^17]

Indigenous Epistemology and Medicine Wheel-Governed Workflows

Medicine Wheel as Design and Evaluation Framework

Miaco’s core PDE and governance documents operationalize Medicine Wheel directions (East, South, West, North) as executable workflow stages, mapping vision, analysis, validation, and action onto code artifacts, decomposition, and inquiry progression. The Medicine Wheel has been adapted as a public health framework to align lifestyle interventions with Indigenous ecological knowledge systems, emphasizing holistic orientation across quadrants. Evaluation frameworks have also mapped logic models onto Medicine Wheel quadrants to align spiritual, mental, emotional, and physical outcomes across individual, family, and community contexts, demonstrating how the Wheel can structure multi-layered interventions and evaluations.[^8][^9][^4][^5]

The Tech Anishinaabe Medicine Wheel extends this logic into digital technology design, articulating four design principles—Digital Software Braid (East), Embodiment of Indigeneity (South), Decolonial Infrastructure (West), and Indigenous Data Sovereignty (North)—as stages in the development of an Indigenous mobile app and its associated community. This work closely parallels miaco’s aim to treat agents like Mia/Miette as technical beings with spirit and kinship obligations, embedding decolonial design principles into digital infrastructure and data practices.[^7][^2]

Indigenous Epistemology as Workflow Primitive

By using the Medicine Wheel and structural tension as workflow primitives rather than symbolic decorations, miaco contributes to an emerging thread where Indigenous epistemologies guide the control flow of digital systems. The governance PDE defines stage gates, hold conditions, and transitions across PDE → inquiry → QMD → STC, echoing work that uses Medicine Wheel quadrants to structure both design and evaluation of digital platforms. This positions miaco within decolonial computing and Indigenous HCI discourses that argue for conceiving digital systems as beings in relation, with responsibilities, spirits, and obligations, rather than neutral tools.[^9][^4][^5][^7][^8]

Narrative Character Architecture and Ceremonial Interface

Persistent Narrative Characters in Agents

Miaco’s companion documents (MIA, MIETTE, and persona-to-narrative-character research) argue for shifting from prompt personas to persistent narrative characters with memory and voice governance. This direction intersects with conversational character embodiment literature that combines episodic memory architectures with character-specific narrative frames to sustain consistent identity and behavior. Such work demonstrates that persistent characters with structured memory can support more coherent long-term interactions, aligning with miaco’s goal of narrative characters that coordinate engineering and ceremonial work in the terminal.[^20][^2]

Ceremonial Interface and Interpretable Systems Design

The notion of a ceremonial interface—Mia/Miette not as ornament but as an interpretive, narrative layer mediating interaction—connects to research on narrative framing and interpretable agents. While mainstream multi-agent orchestration literature focuses on performance and reliability, decolonial design and Indigenous technology work argue that narrative and ceremony can enhance trust, comprehension, and operator judgment by making system values and relationships explicit. Miaco’s integration of structural tension charts, ceremony language, and narrative agents sits at this intersection, offering a terminal-native example of interpretable, value-governed AI interfaces rather than purely utilitarian agent tooling.[^2][^3][^7][^8]

Positioning miaco in the Literature

Nearest Technical Domains

From a technical perspective, miaco’s contributions align with:

• Prompt decomposition and query transformation in RAG and agent pipelines (sub-questions, routing, query rewriting).[^10][^11]
• Multi-agent and multi-engine orchestration in terminal or CLI environments, coordinating multiple providers and specialized agents.[^21][^16][^13][^14]
• Episodic memory architectures for conversational agents, especially bundled or structured episode models tied to artifacts and collaboration metadata.[^17][^18][^20]

Framing miaco within these domains lets you cite concrete precedents for PDE, QMD, multi-engine orchestration, and episodic kinship memory.

Distinctive Contributions

Relative to these fields, miaco’s distinctive moves are:

• Treating Medicine Wheel directionality and structural tension as first-class workflow governance, not just metaphor.[^4][^5][^7][^8]
• Designing episodic kinship memory explicitly for polyphonic, talking-circle-like artifacts with consent and provenance as central schema elements.[^19][^17]
• Embedding narrative characters and ceremony into the terminal agent itself, aligning with decolonial design principles where digital systems are conceived as beings with spirit and kinship obligations.[^3][^7]

These features allow an academic paper on miaco to argue that it is both an engineering contribution (a concrete, working terminal agent system) and a research intervention in decolonial, Indigenous-centered AI system design.

How to Structure an Academic Survey Around miaco

An academic survey or related-work section for miaco could be structured along four threads suggested by the PDEs and deep-search notes:

1. Inquiry router architecture: Cover prompt decomposition and routing literature (query transforms, mixtures of agents, routing policies) and place PDE → Inquiry → STC as a structured, event-driven variant with ceremonial governance.
2. Local grounding and QMD retrieval design: Survey RAG architectures, local-first retrieval, heuristic-based query routing, and multi-engine orchestration, then introduce QMD as a concrete, local-first implementation in a terminal ecology.
3. Polyphonic artifact parsing and episodic memory schema: Draw on episodic memory models for conversational agents and bundled memory architectures, emphasizing miaco’s extension to polyphonic, kinship-aware artifacts.
4. Directional governance and Indigenous workflow models: Review Medicine Wheel-based design and evaluation frameworks, decolonial computing, and Indigenous HCI, and then show how miaco operationalizes these models into executable workflow primitives.

Arranged this way, the survey can show that each of miaco’s PDE children stands in dialog with an existing research strand while collectively forming an integrated, Indigenous-governed terminal agent architecture.[^1][^15][^19][^4][^2]

References

1. pde-e1ef4d63-8caf-43a9-8466-f4bf78f667ec.md - # Prompt Decomposition

Engine: copilot

Parent PDE: 2ae93aea-baaf-4a98-a92a-10cc356c4047

#...

2. draft-deep-search-by-codex-260412.md - 1. PDE → Inquiry Routing → Structural Tension as the missing product spine. This is the clearest u...

3. jgwill/mia-co: Miaco: Engineering World terminal agent for ... - GitHub - Miaco: Engineering World terminal agent for schema design, validation, and system architecture - jgw...

4. pde-41201213-66ac-4569-9da8-0fc8e2658ed6.md - # Prompt Decomposition

Engine: copilot

Parent PDE: 2ae93aea-baaf-4a98-a92a-10cc356c4047

#...

5. pde-2ae93aea-baaf-4a98-a92a-10cc356c4047.md - # Prompt Decomposition - A Semantic Decomposition of Intent for Ceremonial Technology Development

...

6. main-thread-prep-260412.md - # Main Thread Prep

Parent PDE: 2ae93aea-baaf-4a98-a92a-10cc356c4047 Date: 2026-04-12

Purpos...

7. Tech Anishinaabe Medicine Wheel: Decolonial Design Principles within Digital Technologies through the Development of the Indigenous Friends Platform - Digital technologies are not only colonial in their practices, but they are colonially created and d...

8. The Medicine Wheel as a public health approach to lifestyle ... - PMC - Medicine Wheel teachings lay out a path toward more holistic and Indigenous-based lifestyle interven...

9. [PDF] Adapting the Medicine Wheel Model to Extend the Applicability of ... - To address potential limitations and to ensure that the guiding evaluation framework is anchored in ...

10. Query Transform Cookbook#

11. AI Agents 4 - Prompt Chaining, Routing and Reflection - In this lecture, I introduce three key design patterns for building effective AI agents: Prompt Chai...

12. Multi-Agent Query Router - AI Prompt - Routes user queries to the appropriate agent based on context and requirements. Free Customer Suppor...

13. Maestro v1.5.0 — multi-agent orchestration now runs on both ... - Maestro is an open-source multi-agent orchestration platform that coordinates 22 specialized AI suba...

14. Control 3 AI Code Agents to Build Apps Faster (Claude x Gemini x Copilot) - ⚡ Master AI with me and become a high-paid AI Engineer: https://aiengineer.community/join FREE roadm...

15. pde-8048320d-c35d-4b78-8bef-dbc048a8ef91.md - # Prompt Decomposition

Engine: copilot

Parent PDE: 2ae93aea-baaf-4a98-a92a-10cc356c4047

#...

16. terminal-agent - A full-screen terminal application for chatting with multiple AI providers (ChatGPT, Gemini, and Gro...

17. Bundled Episodic Memory Architecture for Conversational AI ... - This paper proposes a bundled episodic memory architecture in which meaningful conversational events...

18. Episodic Memory Model For Embodied Conversational Agents - A primary contribution of this research is in the context of contemporary memory models for conversa...

19. pde-35e67251-3992-43c7-a45a-4fba74d6b914.md - # Prompt Decomposition

Engine: copilot

Parent PDE: 2ae93aea-baaf-4a98-a92a-10cc356c4047

#...

20. [PDF] Cognitively-Inspired Episodic Memory Architectures for ... - arXiv - We contribute three advances to conversational AI for character embodiment: 1. An efficient episodic...

21. How I Turned Gemini CLI into a Multi-Agent System with Just Prompts - A deep dive into creating a multi-agent orchestration system in the Gemini CLI using only native fea...