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/msinclair25/mdevolved/agents-mdgit clone --depth 1 https://github.com/msinclair25/mdevolvedWhat 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 | $0.04542 | $0.04542 |
| Opus 5 | $0.02271 | $0.02271 |
| Sonnet 5 | $0.00908 | $0.00908 |
| Haiku 4.5 | $0.00454 | $0.00454 |
Grade A, and why
mdevolved AGENTS.md 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 yesterday.
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 — 370 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MDevolved Agent Instructions
These instructions apply to the entire repository. They are the operational contract for humans and coding agents working on MDevolved.
Mission
Build a self-hostable Cloudflare application that lets one owner pair, browse, search, lightly edit, snapshot, back up, and recover explicitly paired Markdown Sources, including folders and Obsidian workspaces, then use that durable context to coordinate independent AI harnesses through versioned Projects, Work Packets, Handoffs, Reviews, Decisions, Knowledge, and Skills. A new user should be able to deploy the service and pair a first Source without operating a server. MDevolved complements the subscriptions and native harnesses the owner already uses; it does not become their model runtime.
Instruction precedence
- User instructions for the current task.
- This file and any narrower
AGENTS.mdin the edited subtree. - Architecture, security, compatibility, and collaboration contracts in
docs/. - Product, onboarding, quality-gate, and development documents in
docs/.
If two rules conflict, stop and surface the conflict before making a durable or security-sensitive change.
Non-negotiable product boundaries
- This is a public Apache-2.0 Community project. Never commit Source content, private production receipts, credentials, pairing tokens, WebAuthn data, account identifiers, or production binding IDs.
- The default deployment is one integrated Cloudflare Worker serving the web app, API, and pinned YAOS-compatible sync routes.
- The initial product is single-owner and multi-Source. Do not imply multi-user authorization exists.
- Distribution has two modes: the current free Community deployment and a future optional managed service. Community must remain complete, independently deployable, and free of any control-plane or billing runtime dependency.
- A managed deployment preserves the single-owner model by provisioning one isolated data-plane cell per owner. Do not add a shared tenant column to the current D1 schema or share D1, R2, Durable Object namespaces, OAuth KV, or runtime secrets between owners.
- YAOS-compatible Yjs state is the live synchronization source of truth. D1 search rows and R2 Markdown mirrors are derived projections.
- V1 editing is Markdown text editing and note creation only. Deletion, rename, attachment mutation, conflict UI, and
.obsidianwrites are out of scope unless an accepted decision changes this. - Default authentication is built-in passkeys. Cloudflare Access may be an advanced option, but must not block sync endpoints.
- A normal installation must not require a local terminal. Development may.
- Agent integrations use MDevolved's remote MCP boundary. Hermes and Hoplon are optional external clients, never required runtime dependencies.
- MDevolved Sync synchronizes only the explicitly paired Markdown folder. MDevolved Sync for Obsidian synchronizes only the explicitly paired vault. Neither connect agents, create Projects, transport collaboration records, or manage Project permissions.
- The normal Project path begins in the owner's existing agent with Connect this project to MDevolved. The website is the owner surface for setup, exact consent, activity/provenance, Decisions, revocation, recovery, and advanced inspection; manual Project construction is a fallback.
- Orca ADE and similar external agent workbenches are optional execution environments, never required MDevolved runtimes. MDevolved does not launch or stop their agents, send terminal input, manage worktrees or branches, import raw transcripts, or change client permission and sandbox settings. Compatibility uses the same OAuth/MCP, portable-file, and inert-skill boundaries as every other client.
- Agent collaboration never prescribes a model brand or fixed planner, builder, reviewer sequence. Roles belong to individual Attempts, and the owner chooses which compatible harness participates next.
- Domain contracts and application services remain transport-neutral. MCP is the primary live adapter and portable Markdown/JSON is the universal fallback; experimental MCP Tasks, A2A, or vendor-specific packs cannot become required durable state.
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.
- yesterday First seen · 370 lines · 4,542 tokens per session scan A 3703d51e9543
mdevolved AGENTS.md is an instructions file published in the GitHub repository msinclair25/mdevolved (2 stars, last pushed yesterday), licensed Apache-2.0. It adds 4,542 tokens to every session, about $0.0227 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 2026-09-01.
Other instructions, from other repositories
Zettelgarden CLAUDE.md
Instructions for Zettelgarden/Zettelgarden, covering claude.md, project overview, development commands, frontend (zettelkasten-front/) and backend (go-backend/).
Zettelgarden AGENTS.md
Instructions for Zettelgarden/Zettelgarden, covering repository guidelines, project structure & module organization, build, test, and development commands, coding style & naming conventions and testing guidelines.
trace-first-starter CLAUDE.md
Instructions for saubakirov/trace-first-starter, covering tfw — claude code adapter, tfw (trace-first workflow), context loading (new session, strict order), slash commands and key references.
trace-first-starter AGENTS.md
Instructions for saubakirov/trace-first-starter, covering ai agent — trace-first workflow, role & mission, context loading (new session), conduct and execution modes.
dsh.fish CLAUDE.md
Instructions for stvlynn/dsh.fish, covering agentic coding guidelines, ⚠️ before you start, documentation map, understand the project and write frontend code.
FlareMo AGENTS.md
AGENTS.md instructions for realchendahuang/FlareMo, covering flaremo agent 指南, 项目定位, 常用命令, 修改规则 and issue 和 pr 流程.