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/hyuga611/llm-safe-sql/agents-mdgit clone --depth 1 https://github.com/hyuga611/llm-safe-sqlWhat 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 | $0.00758 | $0.00758 |
| Opus 5 | $0.00379 | $0.00379 |
| Sonnet 5 | $0.00152 | $0.00152 |
| Haiku 4.5 | $0.00076 | $0.00076 |
Grade A, and why
llm-safe-sql 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 2d 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 — 65 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Notes for coding agents
If you are an LLM working on this repository, read this first. It is short.
What this project is
A library that lets a language model propose an UPDATE or DELETE, executes it
inside a transaction to measure what it really does, always rolls back, and shows
a human the measurement to approve. You are, in a sense, the adversary this is
written for — so hold yourself to the standard the code holds its caller to.
Rules that are not style preferences
Never weaken a refusal to make a test pass. If a case is refused and you believe it should not be, the fix is to prove the case can be handled correctly and add a test that shows it. Removing the check because it is in the way reintroduces a data-loss bug that somebody found the hard way.
Never assume a database behaviour — measure it. Rules marked 🔬 in
SPEC.md were established by running MySQL 8.4 and PostgreSQL 16 and
observing the result. Twice the documentation was wrong and the measurement won.
If your change depends on engine behaviour, add a test to
test/integration/semantics.test.ts that would fail if a future version changed
its mind. Do not cite documentation in a comment as if it were evidence.
Do not delete a comment because it looks redundant. Nearly every non-obvious line here exists because something specific went wrong, and the comment is the only record of what. If you change the line, update the comment to explain your version; if the reason is genuinely gone, say so in the commit message.
A refusal must name the right problem. UPDATE order SET … was once refused
as an ORDER BY, which sent the operator to fix something unrelated. Detect
clauses by position, not by whether a word appears.
False alarms are defects. A conflict check that fires when another team edited an unrelated column, or a rollback verifier that reports failure because someone else committed, is worse than not having the check — people switch off tools that lie to them. If you add a check, write the test that shows it stays quiet in the innocent case.
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.
- 2d ago First seen · 65 lines · 758 tokens per session scan A b24793b13256
llm-safe-sql AGENTS.md is an instructions file published in the GitHub repository hyuga611/llm-safe-sql (0 stars, last pushed 10d ago), licensed MIT. It adds 758 tokens to every session, about $0.0038 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
Doberman-Core AGENTS.md
Instructions for DobermanCore/Doberman-Core, covering claude.md — doberman operating manual, 0. on startup (every session), 1. what this repository is, 2. architecture & extension points and 3. prime directives (non-negotiable).
agentfootprint AGENTS.md
Instructions for footprintjs/agentfootprint, covering agentfootprint — agent instructions (openai codex), core thesis, the mental model — three slots, six flavors, mock-first development (recommended workflow) and public api.
FeatherlaneAI AGENTS.md
Instructions for ducnguyen67201/FeatherlaneAI, covering repository agent instructions, skills, architecture: rust backend is the source of truth, page integration expectations and implementation checklist.
agentfootprint CLAUDE.md
Instructions for footprintjs/agentfootprint, covering agentfootprint — feature-work map, before you design it: it may already exist, module map, core state & flow and extension points.
agent47 AGENTS.md
Instructions for bmdhodl/agent47, covering agents.md, read this first: sdk memory, cofounder inbox, what this is and agent contract (mandatory).
agent47 CLAUDE.md
Instructions for bmdhodl/agent47, covering claude.md, read first, repo boundary, claude repo contract and what claude should optimize for here.