Persistent AI memory for growing projects
A reusable template for AI-assisted development. The AI maintains a persistent wiki that tracks both knowledge and codebase context — so it never loses context as your project grows.
AI coding assistants are powerful — but they lose context. This framework fixes that.
As projects grow, AI assistants forget architecture decisions, coding conventions, and prior discussions. Every session starts from scratch.
Knowledge lives in chat logs, comments, and developers' heads. There's no single, structured source of truth the AI can reference.
Without documented conventions and patterns, the AI produces code that conflicts with existing architecture and style.
Two workflows — knowledge ingestion and code changes — both feed the wiki
Drop documents, research, or specs into the raw/ folder for ingestion.
Ask the AI to implement a feature, fix a bug, or refactor — in natural language.
wiki/index.md · wiki/log.md · wiki/overview.md
The wiki compounds over time. Every source ingested and every code change enriches it.
Two disciplines govern every agent action
A single developer agent with modular skills, plus a read-only exploration agent
Handles all code and wiki changes through seven modular skills applied in sequence
Read-only agent for searching the codebase and answering questions — never modifies files or runs commands
| Skill | Purpose |
|---|---|
| Plan | Read requirements, identify affected wiki pages, break work into tasks with verify steps |
| Implement | Write code following conventions, match existing patterns, create tests |
| Test | Run test suite, verify changes, fix failures |
| Wiki Sync | Update wiki pages, run Sync Gate, maintain index/log/overview |
| Review | Lint wiki, check code↔wiki consistency, flag contradictions |
| Clean | Safely delete orphan/deprecated/unused wiki pages |
| Commit | Stage files, write structured commit messages, push to remote |
Skills are applied in sequence during the Post-Change Pipeline — the mandatory 5-step process after every code change.
Everything has a place — sources stay immutable, wiki is AI-maintained
raw/ # Your source documents (immutable)
wiki/ # AI-maintained pages (don't edit manually)
sources/ # Summaries of ingested documents
entities/ # People, orgs, products, tools
concepts/ # Ideas, frameworks, patterns
analyses/ # Comparisons, syntheses
architecture/ # System design, data flows
modules/ # One page per component/service
decisions/ # Architecture Decision Records
conventions/ # Coding standards, project patterns
index.md # Master catalog of all pages
log.md # Chronological operation record
overview.md # High-level synthesis
AGENTS.md # Schema — the single source of truth
scripts/ # Validation and maintenance toolsThree ways to get started — pick the one that fits
Click "Use this template" on the GitHub repo — creates a new repo with the framework pre-loaded.
Use TemplateRun the setup script in your existing project. It creates all directories and files, skipping any that already exist.
# Linux / macOS
curl -sL https://raw.githubusercontent.com/
thongton11314/agent-coding-template/
main/scripts/setup.sh | bash
# Windows (PowerShell)
irm https://raw.githubusercontent.com/
thongton11314/agent-coding-template/
main/scripts/setup.ps1 | iex
In any AI chat inside your project, just say:
"Clone github.com/thongton11314/
agent-coding-template and run
scripts/setup.ps1 (or setup.sh)
to install the AI development
framework into this project."
The AI runs the setup script → framework is installed → AI reads AGENTS.md → ready.
Six platforms are supported out of the box — all config files point to AGENTS.md as the single source of truth
Natural language commands for your AI assistant
Built-in health checks and full extensibility
Run the validation script to catch issues early:
pwsh scripts/validate-wiki.ps1 # Windows
bash scripts/validate-wiki.sh # Linux/macOS
All conventions live in AGENTS.md. Update once — works everywhere.
Add new wiki categories
Change metadata fields
Adjust for your team
Add domain-specific rules
Stop losing context. Start compounding knowledge. Install the AI Development Framework and give your coding assistant permanent memory.