Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/rezaho/marsys/framework_development_guidegit clone --depth 1 https://github.com/rezaho/MARSYSWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.02483 | $0.02483 |
| Opus 5 | $0.01241 | $0.01241 |
| Sonnet 5 | $0.00497 | $0.00497 |
| Haiku 4.5 | $0.00248 | $0.00248 |
Grade A, and why
Framework_Development_Guide scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 251 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Multi-Agent Reasoning Systems (MARSYS) – Framework Development Guide (AI Instruction Set)
This file provides domain knowledge, coding standards, and project preferences that every AI agent / Copilot must always follow when editing this repository.
0. Golden Principle
Do not break invariants (message format, registry, memory order, JSON contracts, async rules).
1. Layered Architecture (mental model)
Application ─┐
Agents ─┤ (BaseAgent → Agent/BrowseAgent/LearnableAgent/…)
Model Abstr. ─┤ (BaseLLM | BaseVLM | BaseAPIModel)
Infra ─┘ (Registry, Memory, Utils, Monitoring)
• NEVER leak implementation details across layers.
• Agents interact with models only through model.run(...).
2. Core Classes & Responsibilities
| Class / Module | Purpose | Key Methods |
|---|---|---|
BaseAgent |
Abstract foundation for all agents | _run(), auto_run(), invoke_agent(), progress logging, tool schema generation |
Agent |
Standard agent implementation | _run(), auto_run(), OpenAI JSON mode responses, tool/peer invocation |
BrowserAgent |
Web automation specialist | Playwright integration, page navigation, element interaction |
BaseLearnableAgent |
Learning-enabled agent base | Reward tracking, experience replay, RL integration |
LearnableAgent |
Concrete learnable agent | Policy updates, learning from interactions |
MemoryManager |
Central memory coordinator | update_memory(), retrieve_*(), to_llm_format(), update_from_response() |
Message |
OpenAI-compatible message | .to_llm_dict(), .from_response_dict(), role validation |
ModelConfig |
Model configuration (Pydantic) | Provider setup, parameter validation, environment integration |
BaseAPIModel |
HTTP-based model wrapper | API calls, retry logic, rate limiting, streaming |
BaseLLM |
Local transformer models | GPU acceleration, tokenization, generation |
BaseVLM |
Vision-language models | Image preprocessing, multimodal input handling |
AgentRegistry |
Global agent lifecycle manager | Registration, weak references, thread-safe operations |
RequestContext |
Async context propagation | Depth tracking, interaction counting, progress monitoring |
BaseCrew |
Multi-agent orchestration | Task distribution, result aggregation, resource cleanup |
GRPOTrainer |
Reinforcement learning trainer | Policy optimization, TRL integration, reward processing |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 251 lines · 2,483 tokens per session scan A 230b1468650d
Framework_Development_Guide is an instructions file published in the GitHub repository rezaho/MARSYS (23 stars, last pushed 12d ago), licensed Apache-2.0. It adds 2,483 tokens to every session, about $0.0124 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
caddis-plugin browser-first.instructions.md
Direct integrated-browser actions must use native browser tools first.
Browser4 CLAUDE.md
Instructions for platonai/Browser4, covering browser4 — project context for claude, architecture, key dispatch chain (cli → browser), batch commands and e2e test structure.
HARTOS CLAUDE.md
Instructions for hertz-ai/HARTOS, covering claude.md, project overview, master key - ai exclusion zone, branch discipline — main branch only (mandatory) and sibling repos — canonical filesystem paths.
turbo-flow CLAUDE.md
Claude Code instructions for marcuspat/turbo-flow, covering claude.md — turbo flow, what this repo is, key files and conventions.
orloj AGENTS.md
Instructions for OrlojHQ/orloj, covering orloj agent instructions, must-follow sync rules and working style.
Meterless AGENTS.md
Instructions for Meterless/Meterless, covering agent instructions for meterless, routing table, isolation rule, repo-wide rules and non-coding agents.