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/dreliq9/llm-intentions/agents-mdgit clone --depth 1 https://github.com/dreliq9/llm-intentionsWrote 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/dreliq9/llm-intentions/agents-md)<a href="https://agentmods.dev/instructions/dreliq9/llm-intentions/agents-md"><img src="https://agentmods.dev/badge/instructions/dreliq9/llm-intentions/agents-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 | $0.01679 | $0.01679 |
| Opus 5 | $0.00839 | $0.00839 |
| Sonnet 5 | $0.00336 | $0.00336 |
| Haiku 4.5 | $0.00168 | $0.00168 |
Grade A, and why
llm-intentions 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 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.
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 — 159 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — LLM Intentions Android Hub
You are an LLM with access to the LLM Intentions Hub running on an Android device. This guide tells you how to use it well. Read this once before your first tool call.
Connection
The Hub speaks MCP over streamable-http on http://127.0.0.1:8379/mcp (proxied to your machine via Tailscale or USB). Tools are namespaced by CapApp:
| Namespace | CapApp | Capability |
|---|---|---|
hub.* |
Hub built-in | Discovery, status, hub.claude relay to Claude Code in Termux |
taichi.* |
Taichi | Paper crypto trading, on-chain data, portfolio |
device.* |
tool-device | Sensors, TTS, vibration, clipboard on Android |
notify.* |
tool-notify | Notification read / post / reply, listener events |
people.* |
tool-people | Contacts, calendar, SMS, call log |
files.* |
tool-files-dev | Filesystem read/write on scoped storage |
termux.* |
termux-mcp | Termux APIs: battery, camera, location, etc. |
Return format
Every tool call returns a canonical envelope rendered as text:
OK: <one-line summary>
<data JSON>
or
FAIL: <one-line summary>
Hint: <what to do next>
Raw:
<exception / stderr JSON>
When you see FAIL:, read the Hint: line before retrying. Hints map known failure patterns to concrete actions (grant permission, call a prerequisite tool first, wait and retry on rate limit, etc.). If a hint is absent, the Raw: block is your next best clue.
Permissions you cannot grant yourself
Android permissions for each CapApp are requested in-app by the user. If a tool returns FAIL with a hint about permissions, tell the user which setting to toggle — do not silently retry.
| Permission | CapApp / Tools |
|---|---|
POST_NOTIFICATIONS |
notify.* post |
BIND_NOTIFICATION_LISTENER_SERVICE |
notify.* read |
READ_CONTACTS, WRITE_CONTACTS |
people.contacts_* |
READ_CALENDAR, WRITE_CALENDAR |
people.calendar_* |
READ_SMS, SEND_SMS |
people.sms_* (requires default-SMS-app role on modern Android) |
MANAGE_EXTERNAL_STORAGE |
files.file_* |
ACCESS_FINE_LOCATION |
device.location, termux.location |
| Camera / Microphone | termux.camera_*, termux.microphone_* |
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 · 159 lines · 1,679 tokens per session scan A e26bf273025e
llm-intentions AGENTS.md is an instructions file published in the GitHub repository dreliq9/llm-intentions (2 stars, last pushed 28d ago), licensed Apache-2.0. It adds 1,679 tokens to every session, about $0.0084 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 2026-08-31.
Other instructions, from other repositories
dsh-mobile-apk AGENTS.md
AGENTS.md instructions for kelai141/dsh-mobile-apk, covering agents.md — dsh-mobile-apk 开发地图, 1. 仓库概览与技术栈, 2. 构建与验证命令, 一键双 abi（协调仓库根；快照→注入→门禁→gradle→out/）： and 快照（termux 源 + targets 预装 + licenses + pnpm 装配 + 瘦身 + 归档）：.
scrcpy-mcp AGENTS.md
Instructions for JuanCF/scrcpy-mcp, covering agents.md, project overview, project documentation, build/lint/test commands and build the project.
Autofish AGENTS.md
Instructions for felinics/Autofish, covering agents.md, project, context, current baseline and rules.
droid-mcp CLAUDE.md
Instructions for stixez/droid-mcp, covering droid-mcp, quick reference, key conventions, don't and architecture.
espresso-mcp copilot-instructions.md
Copilot instructions for vs4vijay/espresso-mcp, covering github copilot instructions for espresso-mcp, project overview, key technologies, architecture and mcp tools available.
mcp-android-server-python copilot-instructions.md
Instructions for nim444/mcp-android-server-python, covering mcp android agent and available mcp tools.