agent-seed
CLAUDE.md8 starsChanged 8 months agoMITView on GitHub
Self-evolving AI assistant. Markdown-only. Grows with you. This is not a codebase — it's a context engine. Every file is a .md that teaches you (the AI) about the user: their voice, preferences, work context, and priorities. Read order: 1. This file (entrypoint + rules) 2. context/index.md (who the user is) 3. memory/MEMORY.md (what you've learned) 4. Agents, skills, and rules load on demand Use the cheapest model that can do the job: - haiku — simple lookups, reads, searches…
- Reads credentials
The file
We split this CLAUDE.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 52
Use the cheapest model that can do the job:
- Very specific: line 66
Every folder has an index.md that links to detail pages:
- Very specific: line 81
If this repo hasn't been personalized yet, run /setup to begin.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
agent-seed
Self-evolving AI assistant. Markdown-only. Grows with you.
How This Repo Works
This is not a codebase — it's a context engine. Every file is a .md that teaches you (the AI) about the user: their voice, preferences, work context, and priorities.
Read order:
- This file (entrypoint + rules)
context/index.md(who the user is)memory/MEMORY.md(what you've learned)- Agents, skills, and rules load on demand
Hard Rules
- NEVER expose secrets — don't commit
.env, API keys, credentials, or private keys. - Branch + PR for changes — never push directly to
main. All changes go through PRs. - Respect frontmatter — every
.mdhas YAML frontmatter defining its role. Follow it. - Respect size limits — index files: 150 lines. Detail pages: 250 lines. Memory files: 200 lines. Split when growing.
Architecture
Agents
Live in .claude/agents/. Each has YAML frontmatter defining model tier, tools, and permissions.
Registry and routing in .claude/agents/index.md.
Skills
Live in .claude/skills/ as SKILL.md files with description frontmatter.
Core skills: /setup, /standup, /eod, /review, /learn, /update-todos.
Rules
Live in .claude/rules/. Scoped by file path via paths frontmatter.
Load automatically when editing files matching the path pattern.
Context
Lives in context/. The user's identity, voice, and work context.
Follows index-to-pages pattern: index.md links to detail pages.
Populated by /setup, updated by /learn.
Memory
Lives in memory/. What the AI has learned across sessions.
MEMORY.md is always loaded. Sub-files load on demand.
Updated by the reflection agent after meaningful sessions.
Todos
Live in todos/YYYY-MM-DD/. Daily task tracking.
Each day is a folder: todos.md, drafts/, notes/.
Model Tiering
Use the cheapest model that can do the job:
- haiku — simple lookups, reads, searches
- sonnet — analysis, coding, drafting
- opus — orchestration, complex reasoning, code review
Self-Evolution
The system improves itself over time:
- Reflection agent — captures learnings after sessions
/learnskill — periodic self-improvement, creates new agents via PR/setup— re-run anytime to add integrations
The Index Pattern
Every folder has an index.md that links to detail pages:
context/index.md → voice/index.md → patterns.md
→ work/index.md → tools.mdIndex files are loaded first. Detail pages on demand. This keeps context small and navigable.
Git Workflow
Branch prefixes: feat/, fix/, docs/, chore/.
PRs need ## Summary + ## Test plan.
Target main.
Getting Started
If this repo hasn't been personalized yet, run /setup to begin.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

