nnsight is a Python library for examining and changing the internal values of deep-learning models, such as layer activations, gradients, and intermediate computations. Researchers and developers use it to study model behavior and causal effects in local PyTorch models or remotely through NDIF. The catalogue entry provides instructions for AI agents working with nnsight.
Borrowing it
Nothing to install: this file belongs to ndif-team/nnsight. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/ndif-team/nnsight/main/CLAUDE.mdgit clone --depth 1 https://github.com/ndif-team/nnsightWrote 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/ndif-team/nnsight/claude-md)<a href="https://agentmods.dev/instructions/ndif-team/nnsight/claude-md"><img src="https://agentmods.dev/badge/instructions/ndif-team/nnsight/claude-md/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/instructions/ndif-team/nnsight/claude-md"><img src="https://agentmods.dev/badge/instructions/ndif-team/nnsight/claude-md.svg" alt="Reviewed on agentmods" width="80" 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 | $0.01260 | $0.01260 |
| Opus 5.5 | $0.00504 | $0.00504 |
| Sonnet 5.5 | $0.00252 | $0.00252 |
| Haiku 4.5 | $0.00126 | $0.00126 |
Grade A, and why
nnsight 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 21d 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 — 44 lines — stays where its author put it; the contents beside it link to each section on GitHub.
nnsight — Agent Guide
Start at docs/index.md, the router for every docs section. Find the one page for the task and read it before writing code. Something broke? Start at docs/errors/symptom-index.md; for an exception or warning — including a plain NameError on a name used after a block — look its message up in docs/errors/index.md.
Contributing
- Base PRs on the
devbranch, notmain. - Conventions, the test-to-run table and the commit format: docs/developing/contributing.md. Code style: STYLE.md. Human-facing setup and the docs standard: CONTRIBUTING.md.
- Docs change in the same PR as the code that invalidates them.
By task
| I want to... | Read |
|---|---|
| Read or modify activations on one forward | usage/trace.md, usage/access-and-modify.md, usage/save.md |
| Generate tokens, or run the task pipeline | usage/generate.md, usage/pipe.md |
| Intervene at specific generation steps | usage/iteration.md |
| Run several prompts at once, or share values between them | usage/invoke-and-batching.md, usage/barrier.md |
| Bundle several traces / share values across them | usage/session.md |
| Reach operations inside a module's forward | usage/source.md |
| Cache activations from many modules | usage/cache.md |
| Take gradients | usage/backward-and-grad.md |
| Check shapes without running the model | usage/scan.md |
| Make persistent edits; skip a module; stop early | usage/edit.md, usage/skip.md, usage/stop-and-early-exit.md |
Pick a model class (HF, any nn.Module, diffusers, vLLM) |
models/index.md |
| Fit a model too big for one GPU | models/tensor-parallel.md, models/quantization.md |
| Run on NDIF | remote/index.md, remote/remote-trace.md |
| Ablate or sweep attention heads | patterns/per-head-attention.md, patterns/ablation.md |
| Sweep many prompts (local or NDIF) | patterns/remote-dataset-sweep.md |
| Run a research technique (logit lens, patching, ablation, steering, probing, SAEs) | patterns/index.md |
| Port code written for nnsight before 0.8 | reference/migrating-to-0.8.md |
Look up an exact API name or a CONFIG setting |
reference/api-quick-reference.md, reference/config.md |
| Extend nnsight (custom model, served value, runtime, backend) | developing/extending.md, developing/extending-envoy.md, developing/index.md |
| Understand why it behaves this way | concepts/index.md |
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.
- 21d ago Changed · -118 lines · -2,455 tokens per session fcc89fbd7941
- 1mo ago Changed · -38 lines · -113 tokens per session 0fffed2094f3
- 2mo ago First seen · 200 lines · 3,828 tokens per session scan A ba4b3c093715
nnsight CLAUDE.md is an instructions file published in the GitHub repository ndif-team/nnsight (1,142 stars, last pushed yesterday), licensed MIT. It adds 1,260 tokens to every session, about $0.0050 per session on Opus 5.5. 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.
nnsight CLAUDE.md compared with instructions from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| nnsight CLAUDE.mdndif-team/nnsight | Instructions file | 1.1k | yesterday | MIT | A | 1,260 | Claude Code |
| vscode buildNext.instructions.mdmicrosoft/vscode | Instructions file | 195k | yesterday | MIT | A | 9,719 | GitHub Copilot |
| next.js AGENTS.mdvercel/next.js | Instructions file | 143k | today | MIT | A | 7,461 | Codex, OpenCode |
| langchain AGENTS.mdlangchain-ai/langchain | Instructions file | 149k | yesterday | MIT | A | 4,680 | Codex, OpenCode |
| vscode oss-third-party-notices.instructions.mdmicrosoft/vscode | Instructions file | 195k | yesterday | MIT | A | 5,516 | GitHub Copilot |
| vscode copilot-instructions.mdmicrosoft/vscode | Instructions file | 195k | yesterday | MIT | A | 3,093 | GitHub Copilot |
| deepseek-harness AGENTS.mddeepseek-ai/deepseek-harness | Instructions file | 274k | 2d ago | MIT | A | 4,313 | Codex, OpenCode |