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/stipend-sh/stipend/agents-mdgit clone --depth 1 https://github.com/stipend-sh/stipendWrote 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/stipend-sh/stipend/agents-md)<a href="https://agentmods.dev/instructions/stipend-sh/stipend/agents-md"><img src="https://agentmods.dev/badge/instructions/stipend-sh/stipend/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 | $0.00561 | $0.00561 |
| Opus 5 | $0.00280 | $0.00280 |
| Sonnet 5 | $0.00112 | $0.00112 |
| Haiku 4.5 | $0.00056 | $0.00056 |
Grade A, and why
stipend 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 — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
Notes for an agent working on this repository. Read this before changing code.
What this is
stipend is a non-custodial USDC wallet on Base that an AI agent installs
without a human. The key is generated on the machine that runs it and never
leaves. Spending limits are enforced in the signing path.
Python only. One runtime dependency: eth-account.
Run it
pip install -r requirements.txt
python -m stipend --help
python -m stipend mcp # MCP server, JSON-RPC over stdio
Docker, for registries that introspect the MCP server:
docker build -t stipend .
docker run -i --rm stipend
The rule that matters
Limits are enforced between deciding and signing, in code the model cannot
reach. policy.check() runs before any signature exists. If you find yourself
moving a check after sign_transfer, or making a limit configurable by anything
that arrives over the network, stop — that is the product, not an
implementation detail.
Specifically, do not:
- widen a cap, add a bypass flag, or make a limit readable-and-writable by a tool exposed over MCP
- move key material anywhere other than the local keystore
- add a hosted mode for the MCP server; a hosted wallet server would put the key on someone else's machine and make this custodial
- weaken
locks.payment_lock()— it is what stops two concurrent sends from both passing the same daily cap
Where the money paths are
policy.py the caps, the allowlist, the confirmation rules
approvals.py one-shot approval tokens, matched exactly, never by ceiling
locks.py OS-level exclusive lock around the whole send
relay.py signing, broadcast, and the durable pending record
ledger.py what happened, and destination trust state
mcp.py the tool surface; deliberately exposes no config writes
Anything touching those four files is a money path. Treat a change there as requiring proof it still refuses what it refused before.
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 · 67 lines · 561 tokens per session scan A 76c554e1d842
stipend AGENTS.md is an instructions file published in the GitHub repository stipend-sh/stipend (0 stars, last pushed 2d ago), licensed Apache-2.0. It adds 561 tokens to every session, about $0.0028 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
agentpay AGENTS.md
Instructions for ANVEAI/agentpay, covering agents.md — agentpay runbook for coding agents, 1. configure (programmatic), create a merchant project + api key, → { project: {...}, apikey: "aplive…" } and add (provision) a paying agent.
agentpay-guard CLAUDE.md
Instructions for theMobiusStrip/agentpay-guard, a project described as: Fail-closed spend guard for x402 agent payments: stateful policy plugin, EIP-3009 replay middleware, and DrainBench — a reproducible adversarial drain benchmark.
swapper-toolkit CLAUDE.md
Instructions for swapperfinance/swapper-toolkit, covering swapper ai agent toolkit, what is swapper finance?, repository structure, how the deposit skill works and deeplink format.
x402-stellar-kit AGENTS.md
AGENTS.md instructions for CodeStrux/x402-stellar-kit, covering working with x402-stellar-kit, the one invariant, production seams, bootstrapping accounts across a custody boundary and add x402 to an existing api.
aegis-protocol CLAUDE.md
Instructions for im-sham/aegis-protocol, covering aegis protocol — claude code project context, what this is, core architecture, job state machine and erc-8004 integration (critical — this is our moat).
spendveto CLAUDE.md
Claude Code instructions for revanthrajeev/spendveto, covering claude.md — working on spendveto, the one rule that governs everything, commands & ports, hard-won gotchas and map.