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/hexproofdev/oring/claude-mdgit clone --depth 1 https://github.com/hexproofdev/oringWhat 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 | $0.01181 | $0.01181 |
| Opus 5 | $0.00590 | $0.00590 |
| Sonnet 5 | $0.00236 | $0.00236 |
| Haiku 4.5 | $0.00118 | $0.00118 |
Grade A, and why
oring CLAUDE.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 yesterday.
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 — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
oring: an unprivileged, self-applied, pre-exec confinement wrapper. It applies
Landlock (filesystem tiers, per-direction TCP, IPC scope) and a seccomp-bpf
denylist to itself, sets NO_NEW_PRIVS, scrubs FDs and signal state, then
execves the target — which inherits every restriction irrevocably and can
never widen it. x86_64 Linux only, enforced by a compile_error! (the seccomp
arch guard is x86_64-specific). Static-musl, no root, no daemon.
Read docs/DESIGN.md before changing anything: it carries the threat model, the
two rings, the seal order, and the residuals oring cannot close. v1.
The discipline
- Fail-closed is the whole posture. Any error before the seal means the target
is never executed, and
panic = "abort"keeps that true even under a panic. Every step inpipeline::seal()is HARD (?); no branch skips a ring, none execs while under-confined. When you touch the seal, keep it that way. - The seal ORDER is load-bearing — a wrong order is a confinement bypass, not a style nit. NNP before seccomp; Landlock before seccomp (its setup syscalls run unfiltered); fd-scrub after Landlock (so a leaked ruleset fd is caught). Know why each step precedes the next (DESIGN §6) before you reorder.
- Verification is the deliverable, not a courtesy. Invariants are proven
(Kani:
x32_guard_total,abi_from_version_clamp—make verify), gated (z3 SMT over the emitted cBPF:tests/seccomp_smt.rs), propertied (proptest), and witnessed on a real Landlock kernel (the containment matrix). A new guarantee needs a runnable check; a property you can only state in prose is an open question, not a fact — say so and open it. - Landlock unions rights along the path walk: access is granted if at least one rule encountered on the path allows it, and there is no deny rule. It is NOT most-nested-wins, and a subpath can never be carved out of a grant. Reason about tiers that way, or you will grant more than you meant to.
- Record rationale where the code is. When you reject an approach, record the rejection; cuts that lose their context get re-proposed.
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.
- yesterday First seen · 71 lines · 1,181 tokens per session scan A e6b95a86c207
oring CLAUDE.md is an instructions file published in the GitHub repository hexproofdev/oring (2 stars, last pushed 1mo ago), licensed BSD-3-Clause. It adds 1,181 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 2026-08-31.
Other instructions, from other repositories
axern AGENTS.md
Instructions for cofy-x/axern, covering axern agent contract, scope, minimum context, sources of truth and repository constraints.
agent-box AGENTS.md
Instructions for 0xferrous/agent-box, covering pre commit, image, returns: /nix/store/ -source, wrappers and portal adr.
bunkervm CLAUDE.md
Claude Code instructions for ashishgituser/bunkervm, covering bunkervm, two operating modes (pervasive throughout the codebase), key files, record / replay / restore / diff and tests / ci.
bunkervm copilot-instructions.md
Copilot instructions for ashishgituser/bunkervm, covering bunkervm — copilot instructions, project overview, architecture, key conventions and adding a new framework integration.
agent-jail AGENTS.md
AGENTS.md instructions for yeet-src/agent-jail, covering building yeet dashboards, mental model, build bottom-up: data → component → layout, 1. get the data right first, in isolation and 2. build each component against a fake signal.
agentos CLAUDE.md
Instructions for rivet-dev/agentos, covering agentos, boundaries, security model, sqlite schema ownership and runtime and registry.