Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Production-grade agent directives for autonomous coding agents: Claude Code, Codex, Cursor, Windsurf, Aider.
| Date | Stars |
|---|---|
| 2026-07-31 | 960 |
| 2026-08-06 | 960 |
| 2026-08-11 | 965 |
| 2026-08-18 | 967 |
| 2026-08-25 | 967 |
| 2026-08-31 | 966 |
| 2026-09-08 | 967 |
| 2026-09-12 | 968 |
| 2026-09-15 | 967 |
| 2026-09-16 | 968 |
| 2026-09-21 | 968 |
| 2026-09-22 | 967 |
| 2026-09-24 | 968 |
| 2026-10-03 | 967 |
| 2026-10-04 | 968 |
| 2026-10-05 | 969 |
| 2026-10-08 | 968 |
| 2026-10-14 | 967 |
| 2026-10-15 | 966 |
| 2026-10-16 | 967 |
| 2026-10-17 | 966 |
| 2026-10-20 | 967 |
| 2026-10-21 | 965 |
| 2026-10-25 | 964 |
| 2026-10-26 | 962 |
| 2026-10-27 | 961 |
| 2026-10-31 | 962 |
| 2026-11-02 | 963 |
| 2026-11-04 | 963 |
Today
— stars today
This week
+2 stars this week
This month
-6 stars this month
Momentum
0.0
growth rate 0.21%/day
# agent-md
Portable contracts for coding agents.
`agent-md` installs one source-of-truth rules file, repo-local hooks,
persistent task state, and a few helper scripts so agents can stop
guessing and start proving their work.
Honest scope:
- Markdown rules are advisory. The agent has to read and follow them.
- Hooks and git hooks are enforceable where the host agent supports them.
- Tests, type-checks, screenshots, and evidence notes are stronger than
model self-assessment.
## Quickstart
```bash
# From inside your project directory
curl -sL https://raw.githubusercontent.com/iamfakeguru/agent-md/main/install.sh | bash
```
Installs support for Claude Code, Codex, Cursor, and Windsurf by default.
## What You Get
```text
your-project/
AGENT.md # source of truth
AGENTS.md # Codex / Cursor / Windsurf
CLAUDE.md # Claude Code wrapper — imports AGENT.md via @AGENT.md (see #3)
agent-md.toml.example # deterministic verification config
.claude/
settings.json
hooks/ # Claude Code enforcement
.codex/
hooks.json
hooks/ # Codex hook wrappers
.agents/skills/ # native Codex skills
agent-md-verify/
visual-evidence/
.cursor/rules/agent-md.mdc # Cursor project rule
.windsurf/rules/agent-md.md # Windsurf workspace rule
.agent-md/
bin/
discover_helpers.sh
doctor.sh
playwright-capture.sh
memory/
agents.md
plan.md
progress.md
verify.md
gotchas.md
.githooks/pre-commit # optional fallback for any agent
```
## The Core Idea
Agent guidance has two layers:
| Layer | Purpose | Reliability |
|---|---|---|
| Rules files | Judgment, planning, style, process | Advisory |
| Hooks/artifacts | Type-checks, tests, lint, state updates, visual evidence | Enforceable where supported |
If something can be forgotten or rationalized away, move it out of prose
and into a checked artifact.
## Runtime Lessons Applied
The production-agent lessons that fit this repo are applied as contracts,
not as copied API boilerplate:
- **Cost/context discipline** — concise directives, helper discovery, and
durable `memory/` files instead of giant repeated prompts.
- **Reliability** — structured hook JSON, deterministic checks,
destructive-command blocks, and explicit unverified-state warnings.
- **Performance** — bounded work slices, selective context loading, safe
parallel tool use, and truncation warnings.
- **Tool use** — structured tool-result guidance, validation before
execution, and clear helper boundaries.
- **Output quality** — tests, runtime evidence, visual artifacts, and
independent/adversarial verification.
API-specific features such as prompt caching, streaming display,
provider retries, idempotency keys, temperature tuning, and batch
processing belong in the application or host runtime. `agent-md` tells
the coding agent to document and verify those choices when the project
uses them; it does not pretend to enforce provider behavior from a rules
file.
## Enforcement Matrix
| Check | Claude Code | Codex | Cursor / Windsurf / Other |
|---|---|---|---|
| Bash safety | Hard block via `.claude/hooks/block-destructive.sh` | Hard block via `.codex/hooks/pre-tool-use.sh` | Not covered |
| Type-check/lint/tests at finish | Hard block via `stop-verify.sh` | Continuation via `.codex/hooks/stop.sh` | Optional `.githooks/pre-commit` |
| `memory/progress.md` updated | Hard block via `state-enforcement.sh` | Continuation via `.codex/hooks/stop.sh` | Optional `.githooks/pre-commit` |
| UI visual evidence | Advisory by default, hard block when `[visual].required = true` | Same through Codex Stop wrapper | Advisory through rules |
| New export without nearby test | Advisory | Advisory through rules/skills | Advisory through rules |
| Truncated Bash output | Advisory | Advisory through Codex PostToolUsExcerpt of 8,049 characters
Read on GitHub
15
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:d4f231c3d4bee0dd, llm:Repository description: 'Production-grade agent directives for autonomous coding agents: Claude Code, Codex, Cursor, Windsurf, Aider.' (no topics, Shell language)
matched fp:d4f231c3d4bee0dd, llm:Repository description: 'Production-grade agent directives for autonomous coding agents: Claude Code, Codex, Cursor, Windsurf, Aider.' (no topics, Shell language)
matched fp:d4f231c3d4bee0dd, llm:Repository description: 'Production-grade agent directives for autonomous coding agents: Claude Code, Codex, Cursor, Windsurf, Aider.' (no topics, Shell language)