93 lines
3.2 KiB
Markdown
93 lines
3.2 KiB
Markdown
# AGENTS.md
|
|
|
|
Operational instructions for AI agents working in this repository.
|
|
|
|
---
|
|
|
|
## Purpose of This Repo
|
|
|
|
`llm-tools` is a **navigation phonebook** for AI agents. It answers the fundamental questions any LLM needs when starting work in Jacob Mathison's environment:
|
|
|
|
- Where is the code? (Gitea, GitHub)
|
|
- What cloud resources exist? (AWS)
|
|
- Which agent am I, and how should I behave? (`agents/`)
|
|
- What workflows are available? (`workflows/`)
|
|
|
|
Read `README.md` first for the full picture.
|
|
|
|
---
|
|
|
|
## Repository Structure
|
|
|
|
```
|
|
llm-tools/
|
|
├── AGENTS.md This file — instructions for AI agents
|
|
├── README.md Overview, quick-start, environment at a glance
|
|
├── llm.txt Plain-text summary for rapid LLM ingestion
|
|
├── manifest.json Machine-readable index of everything in this repo
|
|
│
|
|
├── tools/ Access guides for third-party services
|
|
│ ├── README.md
|
|
│ ├── gitea.llm.md Local self-hosted Gitea
|
|
│ ├── github.llm.md GitHub (MathisonProjects + silmaai)
|
|
│ └── aws-cli.llm.md AWS account and services
|
|
│
|
|
├── agents/ Per-agent configuration and notes
|
|
│ ├── README.md
|
|
│ ├── claude.md
|
|
│ ├── cursor.md
|
|
│ └── codex.md
|
|
│
|
|
├── workflows/ Reusable automation and prompt chains
|
|
│ └── README.md
|
|
│
|
|
├── documents/ Project specs, architecture docs, written context
|
|
│ └── README.md
|
|
│
|
|
└── analytics/ Data sources, dashboards, reporting guides
|
|
└── README.md
|
|
```
|
|
|
|
---
|
|
|
|
## Rules for AI Agents Working Here
|
|
|
|
### Reading
|
|
- Start with `README.md` + `manifest.json` to orient yourself
|
|
- Read the relevant `tools/` guide before touching any external service
|
|
- Check `agents/<your-agent>.md` for any agent-specific config or instructions
|
|
|
|
### Writing
|
|
- **Always update `README.md` and `manifest.json`** when adding files or folders
|
|
- Follow the existing file naming pattern: `<service>.llm.md` in `tools/`
|
|
- Use [Conventional Commits](https://www.conventionalcommits.org/): `docs:`, `feat:`, `chore:`
|
|
- One concern per file — don't bundle unrelated guides together
|
|
|
|
### What Goes Where
|
|
|
|
| Content type | Folder |
|
|
|---|---|
|
|
| External service access guide | `tools/` |
|
|
| Per-agent config / instructions | `agents/` |
|
|
| Reusable prompt chains / automation | `workflows/` |
|
|
| Specs, architecture, written context | `documents/` |
|
|
| Data sources, dashboards, reporting | `analytics/` |
|
|
| Client engagement context | `clients/` |
|
|
| Internal / personal project context | `projects/` |
|
|
|
|
### What NOT to Do
|
|
- Do not commit secrets, tokens, or credentials — reference `credentials/secrets.env` by path only
|
|
- Do not delete files without updating `manifest.json` and `README.md`
|
|
- Do not make destructive AWS or git operations without confirming intent
|
|
|
|
---
|
|
|
|
## Keeping the Index Current
|
|
|
|
Every time a file is added, moved, or removed:
|
|
1. Update `manifest.json` — add/remove the entry
|
|
2. Update `README.md` — update the structure tree if needed
|
|
3. Update the relevant folder's `README.md` if it has a file table
|
|
|
|
This keeps the repo self-describing and useful to the next agent that opens it cold.
|