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 skills add zjunlp/Mechanist --skill zennit-crpgit clone --depth 1 https://github.com/zjunlp/MechanistWrote 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/skills/zjunlp/mechanist/zennit-crp)<a href="https://agentmods.dev/skills/zjunlp/mechanist/zennit-crp"><img src="https://agentmods.dev/badge/skills/zjunlp/mechanist/zennit-crp/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/skills/zjunlp/mechanist/zennit-crp"><img src="https://agentmods.dev/badge/skills/zjunlp/mechanist/zennit-crp.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium analysis-evasion · line 1 Suspicious Unicode normalization or mixed-script contentFix: Review the flagged content for security risks. Ensure no credentials, secrets, or sensitive data are exposed.
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 | $0.00061 | $0.08167 |
| Opus 5 | $0.00030 | $0.04084 |
| Sonnet 5 | $0.00012 | $0.01633 |
| Haiku 4.5 | $0.00006 | $0.00817 |
Grade A, and why
zennit-crp 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 10d 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 — 848 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Zennit-CRP: Concept Relevance Propagation Skill
When to Use
Activate this skill when:
- Generating explainability heatmaps for deep learning model predictions
- Computing concept-conditional attributions (CRP) to understand which latent features drive predictions
- Performing Relevance Maximization (RelMax) or Activation Maximization (ActMax) to find representative samples for neuron/channel concepts
- Building attribution graphs to trace relevance flows through network layers
- Analyzing channel-level importance in convolutional or linear layers
- Comparing what concepts a model uses vs. what activates a neuron (relevance vs. activation)
- Working with LRP (Layer-wise Relevance Propagation) via zennit composites
- Any task involving keywords: CRP, LRP, XAI, explainable AI, feature visualization, relevance propagation, concept attribution, channel relevance, RelMax, ActMax
Quick Reference
- Paper: Nature Machine Intelligence (Open Access)
- PyPI: https://pypi.org/project/zennit-crp/
- GitHub: https://github.com/rachtibat/zennit-crp
- Tutorials: https://github.com/rachtibat/zennit-crp/tree/master/tutorials
- Zennit (dependency): https://github.com/chr5tphr/zennit
- Citation: Achtibat et al., "From attribution maps to human-understandable explanations through Concept Relevance Propagation", Nature Machine Intelligence, 2023
Installation / Setup
Prerequisites
- Python 3.7+
- PyTorch (with CUDA support recommended for large datasets)
- A working model in PyTorch (
torch.nn.Module)
Install from PyPI (recommended, includes fast image utilities)
pip install zennit-crp[fast_img]
Install from Source (to access tutorials)
git clone https://github.com/rachtibat/zennit-crp
pip install ./zennit-crp
Install zennit separately if needed
pip install zennit
Core Features
- Conditional Attributions (CRP): Generate concept-conditional relevance heatmaps by masking relevance flows during backpropagation, isolating the contribution of individual latent concepts to the model output.
- ChannelConcept: Define each channel in Conv2D/Linear layers as a distinct concept with built-in relative importance scoring.
- Custom Concepts: Extend the abstract
Conceptclass to define your own notion of what constitutes a concept (beyond channels). - Relevance Maximization (RelMax): Select dataset samples that maximize the relevance criterion for a given neuron/channel — a more faithful alternative to Activation Maximization (ActMax).
- Activation Maximization (ActMax): Classic approach: select samples that most strongly activate a given unit.
- Feature Visualization: The
FeatureVisualizationclass precomputes and caches reference images (RelMax/ActMax) for all concepts across all layers on a full dataset. - Attribution Graph: Trace relevance as a directed graph through model layers to understand information flow from output back to input.
- Model Graph Analysis: Automatically parse PyTorch JIT-traced model graphs to determine layer connectivity for attribution routing.
- LRP Integration: Built on top of zennit composites (e.g.,
EpsilonPlusFlat,EpsilonAlpha2Beta1) and canonizers (e.g.,SequentialMergeBatchNorm). - Caching System:
ImageCacheand abstractCacheclasses for efficient storage and retrieval of reference images. - Image Utilities:
plot_gridand related helpers incrp.imagefor visualizing attribution results and reference image grids. - Statistics Module: Tools in
crp.statisticsfor aggregating concept relevance statistics across datasets.
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 10d ago First seen · 848 lines · 61 tokens per session scan A 04e250034055
zennit-crp is a skill published in the GitHub repository zjunlp/Mechanist (74 stars, last pushed 14d ago), licensed MIT. It adds 61 tokens to every session and 8,167 once invoked, about $0.0003 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 skills, from other repositories
nanoresearch-writing
Draft a LaTeX research paper from all previous stage outputs.
nanoresearch-experiment
Generate a Python code skeleton from an experiment blueprint.
nanoresearch-ideation
Search academic literature and generate research hypotheses.
nnsight-remote-interpretability
Provides guidance for interpreting and manipulating neural network internals using nnsight with optional NDIF remote execution. Use when needing to run interpretability experiments on massive models (70B+) without local GPU resources, or when working with any PyTorch architecture.
sparse-autoencoder-training
Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.
transformer-lens-interpretability
Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.