Agentic Substrate workflow
Agentic Substrate (claude-user-memory)
A short reversible rule set for an agent: research the unclear, plan proportionally, verify the outcome and keep checked knowledge
- 214 ★ on the repository
- Last commit
Extended access
Why this level
- The agent edits the project instruction file and creates the `.agentic-substrate/` directory, and in an environment without native memory it may keep a notes journal, so backups and confirmation are needed up front
Install
- Works in
Manual install
Read INSTALL.md and install its minimal Agentic Substrate setup in this project. Adapt it to this tool and operating system, preserve my existing configuration, verify the result, and provide an undo procedure. Keep my current model, provider, and memory settings. Ask before adding persistent memory or making changes outside this project.
Give the agent INSTALL.md from the repository or its raw link and paste this text. For repeatability use the document from a reviewed commit.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent edits the project instruction file and creates the `.agentic-substrate/` directory, and in an environment without native memory it may keep a notes journal, so backups and confirmation are needed up front
This is third-party code. Review the repository files before installing.
What it does
Agentic Substrate is a portable working protocol that the agent installs for itself from the `INSTALL.md` document. A person hands the file to Claude Code, Codex or another agent and asks it to read it and install the minimal setup in the project. The agent identifies its environment, preserves the current configuration, verifies the result and describes how to undo everything. The model, provider and memory settings stay as they were. The usual result has three parts: a short addition to the project's native instructions (up to 200 words, always in context), a small `.agentic-substrate/CORE.md` file and an installation receipt with backups and removal steps. The working scheme is this: simple tasks go straight through, while complex or unclear ones start with research and a plan. The agent reports observed outcomes, including checks that could not run, and keeps useful decisions with their context. For chats without file access the document describes a manual instruction pack. A Markdown journal inside the project is optional and enabled only with agreement. The earlier collection of prompts, hooks and installers moved to the archive at `archive/2026-09-26-install-md-migration/legacy/` and is no longer supported.
Who it is for. People who use Claude Code or Codex across several projects and want one short protocol instead of long homemade instructions.
Good fit when
- The agent takes on big tasks without a plan and you want rules for when to research and plan and when to just do it
- The project instruction file has grown and you need an addition of 200 words or fewer
- You want to install the rules so that every step can be reverted using the installation receipt
- The agent often reports "done" without checking and you want a report of the checks actually run
- You want to set up two agents, such as Claude Code and Codex, with one document
Not a fit when
- You need a deterministic installer with a compatibility guarantee: the document calls itself guidance for an agent and promises nothing for every environment
- You expect the rules to improve the model or fine-tune it: the project does not change weights and only governs work order and context
- You need verified persistent memory: its persistence was not tested in this migration
- You plan to use the old install scripts from the root: they stay in the archive as a historical reference
Example request
Read INSTALL.md from the VAMFI/claude-user-memory repository and install the minimal Agentic Substrate setup in this project. Preserve my configuration, verify the result and describe how to undo itLimitations
- The status is a portable beta with agent-guided installation. A fresh install was tested in Codex, Hermes has an offline load check, and its live check stopped on a provider balance.
- Other harnesses, Linux, Windows and manual use in a chatbot are described as guidance without end-to-end testing.
- The rule text steers the agent but enforces nothing. No quality gain or universal compatibility is claimed.
- Persistent memory and the journal are not confirmed by testing, and old links to root scripts no longer open.
How to disable. Ask the agent to run the removal procedure from the installation receipt: it removes its addition from the project instructions and the `.agentic-substrate/` directory while keeping your later edits. If there is no receipt, delete the addition and that directory by hand.
FAQ
How do I install Agentic Substrate in Claude Code?
Give the agent `INSTALL.md` and ask it to install the minimal setup in the project while preserving configuration, verifying the result and providing an undo path. For repeatability take the document from a reviewed commit instead of the `main` link and have the agent record the revision.
What exactly appears in the project after installation?
Usually a short addition of 200 words or fewer in the project's native instruction file, the `.agentic-substrate/CORE.md` file and an installation receipt with backups and removal steps.
Does it need a GPU, a database or Docker?
No. The minimal setup needs no GPU, database, Docker, model download or background service. Local model inference is a separate optional component with a hardware check.
Is anything other than Claude Code supported?
The document is written for different agents. The install was tested on Codex and Hermes has offline checks. For other harnesses only adaptation guidance is given, and chatbots without files get a manual instruction pack.
What happened to the old scripts in the repository?
The earlier collection of prompts, hooks and installers for Claude Code is kept in the migration archive but not supported. For a new install use only `INSTALL.md`.
Discussion
No questions or reviews yet.
Related
WorkflowsOne config for Claude Code and Codex in every repo: shared rules, skill packs, second-model review and a destructive-command guard
A template for Claude Code projects: parallel agent swarms, workflow skills, safety hooks and rules in one install.
Nine markdown files turn a Claude Project into a multi-agent context engine with plan checks and an execution trace
Memory for coding agents in plain Markdown: the agent keeps specs, decisions and lessons itself, and each new session starts from them. No vector database
A plugin and kit for Claude Code and Codex: agent working principles, safety hooks, skills and CLAUDE.md templates in one repository
Evolving Agents, director
Evolving Agents (director)
The director plugin for Claude Code compiles CLAUDE.md and memory into linked team rules and puts them before the agent's plan.
README in short
The page presents the project as a portable agent working scheme and sends the reader to `INSTALL.md`. It then covers the outcome, a table of tests by environment with limits, the move of the old collection into an archive and an invitation to contribute testing on other systems.