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/skynetcmd/m3-memory/claude-mdgit clone --depth 1 https://github.com/skynetcmd/m3-memoryWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/skynetcmd/m3-memory/claude-md)<a href="https://agentmods.dev/instructions/skynetcmd/m3-memory/claude-md"><img src="https://agentmods.dev/badge/instructions/skynetcmd/m3-memory/claude-md.svg" alt="Measured on agentmods" height="20"></a>What 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 | $0.01300 | $0.01300 |
| Opus 5 | $0.00650 | $0.00650 |
| Sonnet 5 | $0.00260 | $0.00260 |
| Haiku 4.5 | $0.00130 | $0.00130 |
Grade B, and why
m3-memory CLAUDE.md scanned grade B with 1 finding 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 4d 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
`~/.claude/settings.json` (only `~/.claude.json`, `.mcp.json`, and plugins are How it starts
The opening of the file, as written. The whole thing — 62 lines — stays where its author put it; the contents beside it link to each section on GitHub.
docs/AGENT_INSTRUCTIONS.md
Git Standards
- Commit Messages: Do NOT include "Co-Authored-By" lines in commit messages. Focus on clear, concise descriptions of "why" and "what".
- Pre-push process (MANDATORY, all agents): After cloning, run
python bin/setup_hooks.pyonce to enable the shared pre-push gate. Before any push, the tool-catalog drift check (python bin/check_tool_catalog_drift.py) and bench-data leakage scan must pass. If you changebin/mcp_tool_catalog.py, regenerate the catalog/inventory and update the "N tools" counts in the same change. Full rationale: the "Pre-push process — ALL agents" section indocs/AGENT_INSTRUCTIONS.md. This is enforced by the local hook AND CI, not just convention. - Cutting a release (version bump): The version has ONE source of truth —
pyproject.toml[project].version. Do NOT hand-edit version strings in the derived manifests. Instead: (1) bumppyproject.toml, (2) runpython bin/sync_manifest_versions.py— it writes the new version into every derived manifest (server.json,mcp-server.json,.claude-plugin/plugin.json,.antigravity-plugin/plugin.json; the plugin manifests are served straight frommainto every user's/plugin install), (3) commit all together.tests/test_tool_count_drift.py::test_all_manifests_synced_to_pyproject_version(runs--check) fails the build if you skip step 2, so a stale manifest can't ship. Plugin manifest descriptions say "100+ MCP tools", never an exact count (also test-guarded).
Homecoming Architecture (decoupled roots)
Persistent state is split across three roots so the engine (databases) and the configuration can live and be secured independently:
M3_CONFIG_ROOT— configuration. Default~/.m3/config. Resolution:M3_CONFIG_ROOTenv >M3_MEMORY_ROOT/config>~/.m3/config. Holds.chatlog_config.json,.migrate_config.json,.agent_os_salt.M3_ENGINE_ROOT— databases + runtime state. Default~/.m3/engine. Resolution:M3_ENGINE_ROOTenv >M3_MEMORY_ROOT/engine>~/.m3/engine. Holdsagent_memory.db,agent_chatlog.db, chatlog state/cursor, spill dir.M3_MEMORY_ROOT— the repo/payload (code), and a master override: if set, config and engine derive from it (/config,/engine) unless their own env vars are set.
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.
- 4d ago First seen · 62 lines · 1,300 tokens per session scan B 27cc57e4d8a5
m3-memory CLAUDE.md is an instructions file published in the GitHub repository skynetcmd/m3-memory (23 stars, last pushed 4d ago), licensed Apache-2.0. It adds 1,300 tokens to every session, about $0.0065 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
obsidian-mind AGENTS.md
AGENTS.md instructions for breferrari/obsidian-mind, covering obsidian mind, hooks, commands, memory and reaching this vault from another repo.
Dragon-Brain CLAUDE.md
Instructions for iikarus/Dragon-Brain, covering dragon brain — claude.md, the harness, audit remediation (april–may 2026, complete 2026-05-09), the lie this audit closed and the contract that matters now.
hybridclaw CLAUDE.md
Claude Code instructions for HybridAIOne/hybridclaw: Follow AGENTS.md as the canonical repo instruction set for this repository.
powerbrain CLAUDE.md
Instructions for nuetzliches/powerbrain, covering claude.md — powerbrain context engine, project overview, architecture, directory structure and components and ports.
obsidian-llm-hub AGENTS.md
Instructions for takeshy/obsidian-llm-hub, a project described as: Obsidian plugin for AI chat, workflow automation, and semantic search — supports Gemini, OpenAI, OpenRouter, Grok, Ollama, and CLI backends.
llm-swarm-router CLAUDE.md
Claude Code instructions for matthewdcage/llm-swarm-router, covering claude code: swarm-llm and honcho memory (mcp).