Menú
在 GPU 服务器上为 accepted_index.csv 构建或刷新论文 embedding 缓存，供 resmax-survey 的语义检索使用。
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
Compile a validated ROI-aware ResearchPack into evidence-grounded IdeaCards, lineage, closest-work checks, experiment planning inputs, and structured negative memory.
Normalize external AI/human survey outputs into reproducible, falsifiable, provenance-tracked research assets with bounded local verification.
Review a validated idea portfolio with standard evidence packages, heterogeneous raw reviews, blocker-first aggregation, and pairwise tournament traces.
构建、规范化和验证 AI 顶会/顶刊基础文献库。产出 paper_database/accepted_index.csv，并维护 manifest、review/cache/schema 覆盖率。
Summarize Resmax daily work from the fixed Codex conversation log into the fixed Obsidian work log. Use when the user says "总结今天工作", "今天工作日志", "记录今天工作", "把今天工作写到工作日志", or asks to summarize today's work from .codex/response-summary-log.md.
Initialize a freshly cloned Resmax checkout. Use when Codex needs to guide first-time setup, materialize local env files from templates, collect required API keys or machine config through questions, restore or build paper_database artifacts, explain hard vs soft requirements, and verify the repo is ready for resmax-database, resmax-embedding, or resmax-survey.
| name | resmax-embedding |
| description | 在 GPU 服务器上为 accepted_index.csv 构建或刷新论文 embedding 缓存，供 resmax-survey 的语义检索使用。 |
用于构建或刷新 paper_database/embedding_cache/qwen3_8b.npz。上游是 resmax-database，下游是 resmax-survey。
paper_database/accepted_index.csv 已由 resmax-database 规范化。.localconfig/server.env 配置了 RESMAX_SSH_HOST 等远程 GPU 参数。SKILL_ROOT=.agents/skills/resmax-embedding
source .localconfig/server.env
ssh "$RESMAX_SSH_HOST" nvidia-smi
rsync -avz $SKILL_ROOT/scripts/ "$RESMAX_SSH_HOST:$RESMAX_SSH_REMOTE_DIR/scripts/"
rsync -avz paper_database/accepted_index.csv "$RESMAX_SSH_HOST:$RESMAX_SSH_REMOTE_DIR/accepted_index.csv"
ssh "$RESMAX_SSH_HOST" "source $RESMAX_SSH_CONDA_INIT && conda activate $RESMAX_SSH_CONDA_ENV && \
cd $RESMAX_SSH_REMOTE_DIR && HF_HUB_OFFLINE=1 python3 scripts/build_cache_multigpu.py \
--accepted accepted_index.csv \
--out .embedding_cache/qwen3_8b.npz \
--gpus 0,1,2,3"
rsync -avz "$RESMAX_SSH_HOST:$RESMAX_SSH_REMOTE_DIR/.embedding_cache/qwen3_8b.npz" \
paper_database/embedding_cache/qwen3_8b.npz
python3 .agents/skills/resmax-database/scripts/normalize_database.py \
--csv paper_database/accepted_index.csv \
--manifest paper_database/manifest.json
python3 .agents/skills/resmax-database/scripts/validate_database.py \
--csv paper_database/accepted_index.csv \
--cache paper_database/embedding_cache/qwen3_8b.npz \
--manifest paper_database/manifest.json
RESMAX_SSH_REMOTE_DIR 以 ~ 开头，远程 ssh 命令中不要把它整体包进引号；cd $RESMAX_SSH_REMOTE_DIR 可由远程 shell 展开，cd "$RESMAX_SSH_REMOTE_DIR" 会创建或访问字面量 ~/... 路径。accepted_index.csv 已经通过 validate_database.py，拉回缓存后优先用 normalize_database.py --skip-normalize 刷新 manifest；普通 normalize 会刷新时间戳字段，可能让缓存 meta 中的 accepted_csv_sha256 再次落后。build_cache_multigpu.py 必须在没有新增 paper_id 时仍能重写 .npz metadata；否则 CSV 内容变化但 paper_id 集合不变时，缓存会保持旧 accepted_csv_sha256。.npz 必须包含：
embeddings: float32 [N, D]paper_ids: safe string dtype，不允许 object dtype / pickle-onlymeta: JSON 字符串，至少包含 model_name、dimension、count、accepted_csv_sha256默认目标是覆盖所有 queryable rows：paper_id 非空且摘要不是空值/占位符。旧缓存如果需要 allow_pickle=True 读取，应视为需要重建。
references/legacy_full_manual.mdSECRETS.md 和 .localconfig/README.md