Skip to content

Harness, agents and self-improvement

Items from the discussions that mention this topic, oldest first. Generated 2026-10-28T19:00:09. Current status of features is in feature-status; the owner's source of truth outranks anything said in a discussion.

Decisions

  • 2026-W06 CONFIRMED: User interactions, common patterns, and structured concept states will be persisted in a vector database to enable personalized agent memory. (src-b612a1e5b0 @ 00:10:00)
  • 2026-W40 CONFIRMED: Speaker 1 will serve as the first testing customer for the harness prototype. (src-13fc0f9115)

Plans

  • 2025-W36 PLANNED: Develop a storyboard depicting four tribal archetypes—Lakshana (Raga/passion), Naga, Deva, and Yaksha (Dvesha/aversion)—calming down upon hearing Dhamma. (src-157495739d @ 01:30:00)
  • 2025-W51 PLANNED: Speaker 1 plans to have a shareable version of the AIware system ready after Christmas. (src-cb9cd2e6f3 @ 00:00:00)
  • 2026-W01 PLANNED: Connect with Ven. Sharada and monastic teachers at Maggavihara to model [doctrine term] conditional relations between mental states in the next version. (src-b9713fb12a @ 00:30:00)
  • 2026-W06 PLANNED: [name withheld] and [name withheld] will collaborate to build an automated reel generation agent using templates, style guides, and video model tokens. (src-b612a1e5b0 @ 00:30:00)
  • 2026-W28 PLANNED: Expose the CLI config harness through a local web server bridge to connect with front-end dynamic UI designs. (src-dadb4daf5f @ 00:20:00)
  • 2026-W39 PLANNED: Provide end users a default vanilla harness on Machine World where task-specific AI agents can be dynamically loaded and attached. (src-3b4db21ed4)
  • 2026-W39 PLANNED: Showcase the depth of Machine World, Inner Realm, and Marga Sakaccha together by creating story-driven visual episodes demonstrating real workflow automation. (src-3b4db21ed4)
  • 2026-W40 PLANNED: The harness will be released as open-source before releasing the commercial Memory as a Service offering. (src-13fc0f9115)
  • 2026-W40 PLANNED: The initial target workflow models a triad of teacher, student, and parent roles. (src-13fc0f9115)
  • 2026-W41 PLANNED: Open source the Machine World interface harness so users can load it across CLI, desktop, and web environments. (src-165c55b513)

Proposals

  • 2025-W32 PROPOSED, medium confidence: The development workflow should move toward agentic architecture using context engineering, spec files, and agent-to-agent protocols rather than ad-hoc chat prompting. (src-23e82f4d71 @ 00:20:00)
  • 2025-W33 PROPOSED: Use n8n containerized flow models to automate repetitive development tasks across chained AI agents with specific tooling access. (src-8fd4bc579e @ 00:10:00)
  • 2025-W33 PROPOSED: The team should examine the A to A (agent-to-agent) protocol to govern interactions among multiple autonomous agents. (src-8fd4bc579e @ 00:20:00)
  • 2025-W42 PROPOSED: Expose the Citta-Cetasika website as an MCP server for AI agents to query mental states and enable voice navigation. (src-1380d0e6e4 @ 00:00:00)
  • 2026-W06 PROPOSED: The AI architecture must transition from reactive, human-prompted execution to proactive, autonomous operation inside Docker containers. (src-b612a1e5b0 @ 00:05:00)
  • 2026-W06 PROPOSED: Scheduled cron jobs running every 2 to 10 minutes will autonomously monitor Gmail, filter spam, and alert users only when critical conditions are met. (src-b612a1e5b0 @ 00:05:00)
  • 2026-W06 PROPOSED: The agent system should expose webhook integrations to connect directly with messaging apps like WhatsApp for regular user communication. (src-b612a1e5b0 @ 00:05:00)
  • 2026-W39 PROPOSED, medium confidence: Integrate third-party transport networks and local services via APIs into calendar agents to schedule commutes and pick-ups automatically. (src-3b4db21ed4)
  • 2026-W40 PROPOSED: Agentic swarms should use an outer-loop shared public ledger or message board where each agent records discoveries for others to learn from. (src-13fc0f9115)
  • 2026-W40 PROPOSED: The harness should load on top of OpenRouter to connect flexibly to any required underlying model or API. (src-13fc0f9115)

Action items

  • 2025-W33 PLANNED, medium confidence: Speaker 2 will set up and test an agentic workflow using n8n for their multimedia production pipeline. (src-8fd4bc579e @ 00:20:00)
  • 2025-W42 PLANNED: Hold a weekly two-hour sync call among team members to align on AI workflows and Git repositories. (src-1380d0e6e4 @ 00:35:00)
  • 2025-W50 PLANNED: [name withheld] or [name withheld] will schedule a screen-sharing session to teach [name withheld] basic Git and repository commit workflows. (src-bb1e0e327c @ 00:35:00)
  • 2025-W51 PLANNED: Test the inner page workflow using the released track 'Vali Val' using meaning text from its pinned comment. (src-cb9cd2e6f3 @ 00:30:00)
  • 2026-W06 PLANNED: The development team will test the agent container locally via Docker Compose before hosting on a dedicated domain or cloud server. (src-b612a1e5b0 @ 00:25:00)
  • 2026-W09 PLANNED, medium confidence: Speaker 1 will share resources and links regarding Claude/agent skills and MCP with [name withheld]. (src-49fef1fd41 @ 00:45:00)
  • 2026-W28 PLANNED: Speaker 2 needs to create an agent to automate the setup of trial accounts and synthetic datasets for live testbed testing. (src-dadb4daf5f @ 00:20:00)
  • 2026-W40 PLANNED: The engineering team must build the harness while Speaker 1 completes the cessation memory MCP backend. (src-13fc0f9115)

Features mentioned

  • 2025-W33 PROPOSED: The website should enable cross-navigation between all seven song versions, showing each version's story, audio narrative, visual music patterns, and creation tools. (src-8fd4bc579e @ 00:00:00)
  • 2026-W06 PLANNED: The final product is planned to be voice-only, using an intermediary mini-agent to synthesize concise spoken summaries rather than reading raw text. (src-b612a1e5b0 @ 00:15:00)
  • 2026-W10 IMPLEMENTED: The system retains conversational and project context in memory, allowing users to recollect project artifacts and execute tasks with an agent swarm. (src-c77b42478f @ 00:00:00)
  • 2026-W10 PROPOSED: Agents are coordinated using skills, hooks (execution triggers based on actions or schedules), and commands (workflow procedures like subtitle injection). (src-c77b42478f @ 00:10:00)
  • 2026-W40 IMPLEMENTED: The cessation memory MCP server is being built to provide Memory as a Service and manage context token bloat across swarms. (src-13fc0f9115)
  • 2026-W41 PROPOSED: Machine World interface harness connects connected devices including mobile phones, TVs, laptops, and home appliances with dynamic rendered context. (src-165c55b513)

Technical details

  • 2025-W31 PROPOSED: Development should utilize IDEs supporting AI agents such as Cursor, Windsurf, or VS Code with GitHub Copilot. (src-1a1fced380 @ 00:35:00)
  • 2025-W32 IMPLEMENTED: The developers rolled back the site to a previous working version after encountering issues with recent modifications. (src-23e82f4d71 @ 00:00:00)
  • 2025-W33 PROPOSED, medium confidence: The multi-agent system should achieve over 80% to 90% accuracy once weight adjustments and workflow dynamic fine-tuning are completed. (src-8fd4bc579e @ 00:10:00)
  • 2025-W33 PROPOSED: An AI orchestrator handles multimodal input and decides whether to route requests to text, audio, video, or 3D depth modules. (src-8fd4bc579e @ 00:15:00)
  • 2025-W33 PROPOSED: For cost efficiency or local execution, open-source models running on local Llama can be integrated into the workflow instead of paid token APIs. (src-8fd4bc579e @ 00:20:00)
  • 2025-W33 PROPOSED: Human-in-the-loop intervention should remain available for intermediate steps where AI outputs need manual fine-tuning. (src-8fd4bc579e @ 00:25:00)
  • 2025-W51 PROPOSED: Agent skills will be exposed to other agents using MCP (Model Context Protocol) and Agent-to-Agent (A2A) protocols. (src-cb9cd2e6f3 @ 00:05:00)
  • 2026-W06 CONFIRMED: OpenRouter does not provide video generation models, requiring external API endpoints and tokens for video workflows. (src-b612a1e5b0 @ 00:35:00)
  • 2026-W07 PROPOSED, medium confidence: Use Docker MCP containers or Antigravity MCP tools for standardizing agent tool execution. (src-7d6f99df81 @ 01:00:00)
  • 2026-W09 PROPOSED: The automated agent architecture will combine agent skills and MCP tools to manage audio processing and distribution. (src-49fef1fd41 @ 00:40:00)
  • 2026-W10 CONFIRMED: The system architecture is a native AI backbone where layer-level orchestration manages different AI functions, omni models, and tool calls. (src-c77b42478f @ 00:00:00)
  • 2026-W28 PROPOSED: The output evaluation judge model must be trained on a completely different dataset to prevent bias in self-improvement loops. (src-dadb4daf5f @ 00:15:00)
  • 2026-W39 CONFIRMED: Agent-to-agent transactions using tokens can leverage emerging infrastructure such as Stripe's agent transaction capabilities. (src-3b4db21ed4)
  • 2026-W40 CONFIRMED: Machine World provides a harness to give users configuration capability over long horizon agents. (src-13fc0f9115)
  • 2026-W40 CONFIRMED: The harness allows attaching long-running agents that pass through a Recursive Self-Improvement (RSI) cycle. (src-13fc0f9115)
  • 2026-W40 CONFIRMED: The harness is defined purely declaratively as a description containing a system part and a user part, which a runner dynamically renders into a UI. (src-13fc0f9115)
  • 2026-W41 IMPLEMENTED: Machine World core architecture revolves around a local memory management module that loads workflow memory snapshots into agents. (src-165c55b513)
  • 2026-W41 IMPLEMENTED: Speaker 1 uses cmux on macOS as an MCP-driven terminal tool to orchestrate Claude CLI sessions and agent swarms. (src-165c55b513)
  • 2026-W42 IMPLEMENTED: The memory substrate utilizes isolated memory strips assigned per user to ensure context isolation and persistent multi-session memory. (src-93d91aeca1)
  • 2026-W42 IMPLEMENTED: The backend execution engine runs a Recursive Self-Improvement (RSI) loop optimized for agent creation and iterative fine-tuning. (src-93d91aeca1)