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/floe-labs/floe-guard/agents-mdgit clone --depth 1 https://github.com/Floe-Labs/floe-guardWrote 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/floe-labs/floe-guard/agents-md)<a href="https://agentmods.dev/instructions/floe-labs/floe-guard/agents-md"><img src="https://agentmods.dev/badge/instructions/floe-labs/floe-guard/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 5.1 | $0.01175 | $0.01175 |
| Opus 5 | $0.00588 | $0.00588 |
| Sonnet 5 | $0.00235 | $0.00235 |
| Haiku 4.5 | $0.00118 | $0.00118 |
Grade A, and why
floe-guard 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 6d 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 — 83 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — floe-guard
Instructions for AI coding agents. Most often you're here to add floe-guard to an agent (below); the last section covers working on this repo.
What floe-guard is
It tells you what every AI call really costs. It meters
STT + TTS + LLM + telephony per call, out of the box (Pipecat
and LiveKit, Python & TypeScript), and keeps a live ledger of
your agent's real spend. Connect free (one key) for your Coverage Score and 7-day
history. It also hard-stops the next turn before it crosses your USD ceiling — a
runaway loop dies at $0.10 instead of $4,000. Runs in your process: no account, no
signup, no network, no telemetry. It also guards any LLM agent (OpenAI, Anthropic, CrewAI,
LangChain, …) via plain check() / record() or an adapter. Python (pip) +
TypeScript (npm). MIT. Built by Floe — know what every AI call
really costs: cost per call across every vendor, margin per client, and rebilling your own
customers off the actuals.
Add it to an agent — unattended steps
-
Install
- Python:
pip install floe-guard - TypeScript / Node:
npm i floe-guard
- Python:
-
Wire the hard-stop around each model call (
checkbefore,recordafter):from floe_guard import BudgetGuard guard = BudgetGuard(limit_usd=5.00) # the ceiling guard.check() # before each LLM call — raises BudgetExceeded if it'd cross resp = call_your_llm(...) # your existing call guard.record("gpt-4o", resp.usage.prompt_tokens, resp.usage.completion_tokens)TypeScript is the same shape (
guard.check()/guard.record(...)); seejs/.- Framework adapters (no manual check/record): OpenAI, Anthropic, Gemini, CrewAI, LangChain, LangGraph, LiteLLM, Vercel AI SDK — README §"Framework adapters".
- Voice (per-turn): Python
floe_guard.integrations.pipecat/.livekit; TSfloe-guard/adapters/{livekit,vapi,retell}— reserve-before-turn. - Paid tools:
reserve_tool()/settle_tool()block before the tool runs.
-
Verify offline (no key, no network):
floe-guard demo— the guard halts a stubbed loop after a few iterations. No repo checkout (the wheel does not shipexamples/).
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.
- 6d ago First seen · 83 lines · 1,175 tokens per session scan A 5765405069c0
floe-guard AGENTS.md is an instructions file published in the GitHub repository Floe-Labs/floe-guard (385 stars, last pushed 3d ago), licensed MIT. It adds 1,175 tokens to every session, about $0.0059 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
balaka CLAUDE.md
Claude Code instructions for artivisi/balaka, covering claude instructions, project overview, current status, key files and development guidelines.
expense-budget-tracker AGENTS.md
AGENTS.md instructions for kirill-markin/expense-budget-tracker, covering expense-budget-tracker, rules, components, supported clients and key paths.
hvtracker AGENTS.md
AGENTS.md instructions for YugantM/hvtracker, covering agents.md, hard rules, gates — every pr, all three green, map — grep, don't read wholesale and local dev & checks.
AUTHENSOR AGENTS.md
Instructions for AUTHENSOR/AUTHENSOR, covering agents.md — authensor integration guide for ai agents, what authensor does for you, quick integration, if you have access to http and if you have access to npm / npx.
ardur AGENTS.md
AGENTS.md instructions for ArdurAI/ardur, covering ardur agent instructions, 0. first action in every new session, 1. what ardur is, and what it does not claim, 2. tl;dr commands and 3. architecture and trust boundaries.
ardur CLAUDE.md
Claude Code instructions for ArdurAI/ardur: This repository keeps a single, canonical set of agent instructions in AGENTS.md. This file exists only so that Claude Code loads them automatically; it deliberately holds no rules of its own, so that the two can never drift apart.