The authoritative source for endogenic / agentic product design and development workflows, best practices, agent files, and automation scripts.
Endogenic development is the practice of building AI-assisted systems from the inside out — scaffolding from existing knowledge, encoding operational wisdom as scripts and agents, and letting the system grow intelligently from a morphogenetic seed rather than through vibe-driven prompting.
This repo holds the canonical reference for how we work with AI coding agents. It is a living manifesto, a workflow library, and a scripts catalog — not an application codebase.
Contents:
| Path | Purpose |
|---|---|
MANIFESTO.md | Core philosophy and dogma of endogenic development |
AGENTS.md | Root guidance for all AI coding agents operating in this repo |
CONTRIBUTING.md | How to contribute to this repo |
docs/glossary.md | Canonical definitions for all key project terminology |
.github/agents/ | VS Code Copilot custom agent files |
scripts/ | Reusable automation and utility scripts |
docs/ | Guides, protocols, and best practice documentation |
Scaffold from existing system knowledge — do not author from scratch in isolation. Every new file, agent, or script should derive from what the system already knows about itself.
If you have performed a task twice interactively, the third time is a script. Repeated or automatable tasks must be encoded as committed scripts or automation before being performed again by hand. See docs/guides/programmatic-first.md.
Every change to a workflow, agent, or script must be accompanied by clear documentation. The docs folder is as important as the code.
Prefer running locally. Minimize token usage by encoding knowledge as scripts, caching context, and using local models where possible. See the Running Locally guide.
The .github/agents/ directory contains VS Code Copilot custom agents. To use them:
- Open this repo (or any consuming repo that references these agents) in VS Code
- Open Copilot Chat
- Use
@<agent-name>to invoke any agent in the fleet
See docs/guides/agents.md for the complete guide.
All scripts are invoked via uv run:
# Initialize a scratchpad session file for today
uv run python scripts/prune_scratchpad.py --init
# Start the scratchpad watcher (auto-annotates session files on change)
uv run python scripts/watch_scratchpad.pySee scripts/README.md for the full catalog.
.github/
agents/ # VS Code Copilot custom agent files (.agent.md)
AGENTS.md # Agent authoring rules and conventions
README.md # Agent fleet catalog
executive-scripter.agent.md
executive-automator.agent.md
ISSUE_TEMPLATE/ # GitHub issue templates
pull_request_template.md
docs/
glossary.md # Canonical definitions for all key project terminology
guides/
agents.md # How to author and use agents
programmatic-first.md # The programmatic-first principle
session-management.md # Cross-agent scratchpad and session protocols
local-compute.md # Running locally and reducing token usage
scripts/
README.md # Script catalog
prune_scratchpad.py # Scratchpad session file manager
watch_scratchpad.py # File watcher for auto-annotating session files
AGENTS.md # Root agent guidance (programmatic-first, commit discipline)
CONTRIBUTING.md # Contribution guidelines
MANIFESTO.md # Endogenic development philosophy and dogma
- AccessiTech/EndogenAI — the experimental MCP framework where these patterns were pioneered
- PR #41: Programmatic-First Principle — the defining PR for the programmatic-first workflow
- Issue #35: Running Locally — the priority issue that inspired this repo
Apache 2.0 — see LICENSE.