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 HorizonRobotics/OE-Skills --skill lightcompress-quant-exploregit clone --depth 1 https://github.com/HorizonRobotics/OE-SkillsWrote 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/horizonrobotics/oe-skills/lightcompress-quant-explore)<a href="https://agentmods.dev/skills/horizonrobotics/oe-skills/lightcompress-quant-explore"><img src="https://agentmods.dev/badge/skills/horizonrobotics/oe-skills/lightcompress-quant-explore/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/horizonrobotics/oe-skills/lightcompress-quant-explore"><img src="https://agentmods.dev/badge/skills/horizonrobotics/oe-skills/lightcompress-quant-explore.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 | $0.00042 | $0.01312 |
| Opus 5 | $0.00021 | $0.00656 |
| Sonnet 5 | $0.00008 | $0.00262 |
| Haiku 4.5 | $0.00004 | $0.00131 |
Grade A, and why
lightcompress-quant-explore 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 9d 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 — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
LightCompress 量化探索实验
执行流程（4 阶段）
Phase 1: 准备实验
调用 prepare_experiment.py 创建实验目录和 YAML 配置：
python {skill_dir}/scripts/prepare_experiment.py \
--model /path/to/model \
--model-type Qwen2 \
--method smoothquant \
--w-bit 8 \
--a-bit 8
注意：不指定
--workspace时，脚本默认使用{当前工作目录}/experiments作为输出目录。
脚本输出 JSON 包含：
experiment_dir: 实验目录路径config_path: YAML 配置路径yaml_content: 完整 YAML 配置内容
Phase 2: 用户确认
向用户展示完整的 YAML 配置内容，让用户检查是否符合预期。
使用 AskUserQuestion 询问：开始实验 / 取消
Phase 3: 执行实验
用户确认后，调用 execute_experiment.py 执行实验：
python {skill_dir}/scripts/execute_experiment.py {experiment_dir}
Phase 4: 保存精度缓存
实验成功后，检查 <project_root>/.claude/skills/quant-accuracy-cache/ 是否存在：
- 存在：调用
quant-accuracy-cacheskill 保存结果 - 不存在：跳过
参数说明
prepare_experiment.py 参数
| 参数 | 必需 | 说明 | 默认值 |
|---|---|---|---|
--model |
✅ | 模型路径 | - |
--model-type |
✅ | 模型类型 | - |
--method |
✅ | 量化方法 | - |
--w-bit |
权重量化位宽 | 4 | |
--a-bit |
激活量化位宽 | 8 | |
--fast-mode |
快速验证模式 | false | |
--save-artifacts |
保存量化产物 | false | |
--workspace |
工作目录 | ./experiments | |
--mix-bits |
混合精度组(可多次),见下方示例 | - | |
--special |
覆盖 quant.special 参数(可多次),KEY=VAL |
- | |
--calib-path |
校准数据集路径 | 内置 wikitext2_calib | |
--calib-name |
校准数据集名 | wikitext2 | |
--calib-n-samples |
校准样本数(显式值不被显存降级) | 自动 | |
--calib-seq-len |
校准序列长度 | 模板默认 | |
--calib-preproc |
校准预处理 | wikitext2_gptq | |
--eval-path |
评估数据集路径 | 内置 wikitext2_eval | |
--eval-seq-len |
评估序列长度 | 模板默认 | |
--no-eval-pretrain |
只评 fake_quant,跳过 pretrain | false | |
--inference-per-block |
逐 block 推理评测(大模型省显存) | false | |
--save-path |
fakequant 保存路径(指定即开 save_fake) | - |
混合精度 (mix_bits) 示例
将 q/o_proj + MoE expert 设为 W4、其余 W8(Gemma4Moe 常用):
python {skill_dir}/scripts/prepare_experiment.py \
--model /jfs-public/openexplorer_llm/models/gemma-4-26B-A4B-it \
--model-type Gemma4Moe --method gptq --w-bit 8 --a-bit 8 \
--mix-bits "name=qo_experts_w4;bit=4;layers=self_attn.q_proj,self_attn.o_proj,experts.experts.*.gate_proj,experts.experts.*.up_proj,experts.experts.*.down_proj" \
--calib-n-samples 256 --calib-seq-len 1024 \
--no-eval-pretrain --inference-per-block \
--save-path /path/to/save/ \
--env-activate ~/miniconda3/etc/profile.d/conda.sh
What ships with it
15 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.
- base_template.yml 3.6 KB
- examples.md 4.2 KB
- methods_config.yml 4.0 KB
- pretrain_accuracy_cache.yml 1.3 KB
- reference.md 9.5 KB
- scripts/execute_experiment.py 4.3 KB runs code
- scripts/lib/__init__.py 2.8 KB runs code
- scripts/lib/config.py 32 KB runs code
- scripts/lib/experiment_runner.py 18 KB runs code
- scripts/lib/gpu_selector.py 3.7 KB runs code
- scripts/lib/log_parser.py 9.8 KB runs code
- scripts/lib/report_generator.py 19 KB runs code
- scripts/prepare_experiment.py 14 KB runs code
- scripts/run_lightcompress_exploration.py 16 KB runs code
- scripts/test_modules.py 11 KB runs code
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.
- 9d ago First seen · 116 lines · 42 tokens per session scan A ef83d7504983
lightcompress-quant-explore is a skill published in the GitHub repository HorizonRobotics/OE-Skills (19 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 42 tokens to every session and 1,312 once invoked, about $0.0002 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
agent-platform-rag-engine-management
Manage and query Agent Platform RAG Engine Corpora and retrieve grounded contexts using the Google GenAI SDK. Use when listing RAG corpora or files, inspecting a corpus, retrieving contexts, or generating content grounded in a RAG corpus. Do not use for standard database queries (use SQL/Spanner skills), Google…
agent-platform-model-registry
Agent Platform Model Registry Management. Use when you need to upload, list, describe, update, or delete machine learning models (and their versions) in the Agent Platform Model Registry. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform models.
foundry-config-setup
Resolve missing setup caused by a hardcoded Foundry project endpoint or model in a sample. Use when a sample fails because it uses a placeholder/hardcoded projectendpoint (for example "https://your-project.services.ai.azure.com") or a hardcoded model instead of reading them from the environment.
google-cloud-solution-agentic-analytics-spark-knowledge-catalog
Discovers requirements and generates guidance to design and deploy a governed, secure agentic-analytics solution for data that's distributed across Google Cloud, other cloud providers, or on-premises. Data that's outside Google Cloud (such as data from Databricks, Snowflake, Salesforce, SAP, or Oracle systems) is…
training-check
Interactively monitor training metrics from the current Codex session, periodically checking WandB or fallback logs for NaN, divergence, plateaus, and broken runs.
nemo-automodel-launcher-config
Configure NeMo AutoModel job launches for interactive runs, Slurm clusters, and SkyPilot cloud execution.