amr/memex
1
0
Fork 0
Code Issues 12 Pull Requests Actions Packages Projects 1 Releases Wiki Activity

REFAC: Standardize on Cognitive Cycle and update documentation

Browse Source
This commit is contained in:
Amr Gharbeia
2026-04-14 15:59:19 -04:00
parent 34a210228e
commit e29ca5679e
46 changed files with 195 additions and 268 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
gtd_archive_2026
View File

@@ -13664,7 +13664,7 @@ CLOSED: [2026-04-04 Sat 17:36]
​The Proxmox Migration: Move the agent from the low-power machine into a dedicated LXC container on the new rack.
** ​Phase III: The org-agent Microkernel (The "Native" Phase)
** ​Phase III: The opencortex Microkernel (The "Native" Phase)
​Hollowing out the modern tools in favor of Lisp.
​The Lisp Transition: Replace OpenClaw’s Python logic with your minimalist Common Lisp heartbeat.
@@ -13942,7 +13942,7 @@ Server Rack Assembly: Build the 4080/4040 aluminum skeleton.
Local AI Capability: Install the RTX 6000 Pros. Move from remote APIs to local inference using Ollama/vLLM (DeepSeek-R1 / Qwen3-Coder).
The Proxmox Migration: Move the agent from the low-power machine into a dedicated LXC container on the new rack.
***** Phase III: The org-agent Microkernel (The "Native" Phase)
***** Phase III: The opencortex Microkernel (The "Native" Phase)
:PROPERTIES:
:CREATED: [2026-03-20 Fri 08:20]
:END:
@@ -13956,16 +13956,16 @@ The final evolution into a self-hosted Lisp Machine.
The FPGA Sidecar: Activate the hardware-level "Thalamus" to verify Lisp forms.
Unikernel/Bare Metal: Bootstrap Sol out of the Linux environment and directly onto the hardware, where the Org-mode tree and the Lisp Heap become the operating system.
***** Org-agent
***** OpenCortex
:PROPERTIES:
:CREATED: [2026-03-20 Fri 08:00]
:END:
By formalizing this, you are effectively designing the blueprint for Sol.
To answer your biggest question first: No, org-agent will not replace OpenCode or the Claude Agent SDK—it will act as their orchestrator and wrapper. Tools like OpenCode (which focuses on terminal-based, plan-first coding with AGENTS.md) and the Claude Agent SDK (which provides programmatic access to Anthropic's tool-calling and multi-agent teams) are phenomenal execution engines. However, they are built around Markdown, JSON, and Python/TypeScript ecosystems.
org-agent replaces their management layer. Instead of letting OpenCode or Claude dictate your workflow, org-agent sits above them. When org-agent decides a complex coding task is required, it simply invokes OpenCode or a Claude subagent via a CLI or API call, passing along a compiled .org file as context.
To answer your biggest question first: No, opencortex will not replace OpenCode or the Claude Agent SDK—it will act as their orchestrator and wrapper. Tools like OpenCode (which focuses on terminal-based, plan-first coding with AGENTS.md) and the Claude Agent SDK (which provides programmatic access to Anthropic's tool-calling and multi-agent teams) are phenomenal execution engines. However, they are built around Markdown, JSON, and Python/TypeScript ecosystems.
opencortex replaces their management layer. Instead of letting OpenCode or Claude dictate your workflow, opencortex sits above them. When opencortex decides a complex coding task is required, it simply invokes OpenCode or a Claude subagent via a CLI or API call, passing along a compiled .org file as context.
Here is the developer brief for building the system.
Project Brief: org-agent
Project Brief: opencortex
Mission: Build a hyper-minimalist, self-editing, proactive AI agent framework in Common Lisp. The system must use Org-mode as its exclusive interface for memory, state, and skill configuration, rejecting Markdown and JSON overhead.
Target Environment: A Linux/Proxmox homelab, running local inference servers, heavily utilizing GitOps and Lisp-based development.
1. Core Architecture (The Lisp Microkernel)
@@ -13979,18 +13979,18 @@ Instead of relying on SKILL.md or OpenCode's AGENTS.md, all capabilities are def
* Dynamic Loading: The agent reads these files on boot. If the agent generates a new .org file via an LLM output, it instantly acquires that new skill on the next heartbeat.
* Progressive Disclosure: To save token context, the agent only loads the #+DESCRIPTION of a skill initially. It must explicitly request the full #+BEGIN_SRC implementation if it decides to use it.
3. LLMs and Coding Agents as "Effectors"
org-agent treats AI models and external SDKs exactly like it treats a Proxmox server—as a downstream tool to be managed.
| External Tool | How org-agent Uses It | Trade-off |
opencortex treats AI models and external SDKs exactly like it treats a Proxmox server—as a downstream tool to be managed.
| External Tool | How opencortex Uses It | Trade-off |
|------------------------+-----------------------------------------------------------------------------------------------------------------------------------+---------------------------------------------------------------------|
| Local Inference (vLLM) | A base skill (llm-local.org) containing an HTTP POST request to your RTX 6000 cluster for fast, cheap, private reasoning. | Requires manual mapping of Lisp s-expressions to the local model's API. |
| OpenCode | A wrapper skill (skill-opencode.org) that triggers the opencode CLI to execute a massive refactoring job, returning the git diff to org-agent. | Abandons OpenCode's native TUI in favor of headless execution. |
| OpenCode | A wrapper skill (skill-opencode.org) that triggers the opencode CLI to execute a massive refactoring job, returning the git diff to opencortex. | Abandons OpenCode's native TUI in favor of headless execution. |
| Claude Agent SDK | A wrapper skill that triggers a Python script leveraging Anthropic's SDK to spin up a multi-agent "Team" for deep architectural planning. | Costs external API tokens; requires Python alongside Lisp. |
4. Security & Isolation
Because org-agent runs directly on the host machine to manage Proxmox and GitOps, security is handled via strict allow-lists within the Lisp kernel.
Because opencortex runs directly on the host machine to manage Proxmox and GitOps, security is handled via strict allow-lists within the Lisp kernel.
* The LLM backend is instructed to output strictly formatted Lisp s-expressions (e.g., (invoke-skill "proxmox" "restart-staging")).
* The Lisp Executor will only evaluate forms where the car (the function name) matches a predefined, safe whitelist.
The Verdict on the Workflow
By building org-agent, you are creating a system where your daily GTD notes, your infrastructure commands, and your AI's source code all share the exact same syntax. It is the ultimate manifestation of "Code is Data."
By building opencortex, you are creating a system where your daily GTD notes, your infrastructure commands, and your AI's source code all share the exact same syntax. It is the ultimate manifestation of "Code is Data."
Would you like me to draft the Lisp parser function that reads an .org skill file and translates its #+NAME and #+BEGIN_SRC blocks into callable Lisp functions for the agent's whitelist?
***** Agora