GB-Power-Market-JJ caixu-ingest-materials
Import Personal Documents. Use when the user wants to import a local directory or explicit files into 材序 at the ingest stage, including “导入一批材料”“先把这个目录吃进去”“先建立可继续建库的解析上下文”. Prefer caixu-skill when the user asks for the full end-to-end mainline or is unsure which stage to run. This skill creates or loads a library, starts an ingest pipeline run, lets the agent choose per-file routes over low-level OCR and parser tools, persists ParsedFile records, and hands off library_id plus run_id to build-asset-library. Do not use it for asset extraction, lifecycle judgment, package building, or submission.
install
source · Clone the upstream repo
git clone https://github.com/GeorgeDoors888/GB-Power-Market-JJ
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/GeorgeDoors888/GB-Power-Market-JJ "$T" && mkdir -p ~/.claude/skills && cp -r "$T/openclaw-skills/skills/angjustinl/asset-library-skill/caixu-ingest-materials" ~/.claude/skills/georgedoors888-gb-power-market-jj-caixu-ingest-materials && rm -rf "$T"
OpenClaw · Install into ~/.openclaw/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/GeorgeDoors888/GB-Power-Market-JJ "$T" && mkdir -p ~/.openclaw/skills && cp -r "$T/openclaw-skills/skills/angjustinl/asset-library-skill/caixu-ingest-materials" ~/.openclaw/skills/georgedoors888-gb-power-market-jj-caixu-ingest-materials && rm -rf "$T"
manifest:
openclaw-skills/skills/angjustinl/asset-library-skill/caixu-ingest-materials/SKILL.mdsource content
Import Personal Documents
在用户要“导入一批材料”“先把这个目录吃进去”“先建立可继续建库的解析上下文”时使用这个 skill。
Quick flow
- 创建或加载
,再创建library_id
pipeline runingest - 调
,让 agent 为每个文件选择 routelist_local_files - 调低层工具提取文本,归一化成
,写入ParsedFileupsert_parsed_files
Read next only when needed
- 输入是目录、混合格式、需要 route 判断时,读 references/workflow.md
- 需要确认低层工具输入输出、route 含义或
结构时,读 references/tool-contracts.mdPipelineRun - 需要最小 JSON 输出模板时,读 references/output-patterns.md
- 遇到 route 决策失败、部分文件失败或写库失败时,读 references/failure-modes.md
Required tools
caixu-data-mcp.create_or_load_librarycaixu-data-mcp.create_pipeline_runcaixu-ocr-mcp.list_local_filescaixu-ocr-mcp.read_local_text_filecaixu-ocr-mcp.extract_parser_textcaixu-ocr-mcp.extract_visual_textcaixu-ocr-mcp.render_pdf_pagescaixu-data-mcp.upsert_parsed_filescaixu-data-mcp.append_pipeline_stepcaixu-data-mcp.complete_pipeline_run
Required input
or explicit local filesinput_rootlibrary_id?owner_hint?
Workflow
- 如果上下文里没有
,先调library_id
。create_or_load_library - 创建
run;之后每个关键动作都要追加 step。ingest - 先用
展开目录,再让 agent 为每个文件选择list_local_files
。text | parser_lite | parser_export | ocr | vlm | skip - 只调低层工具,不自己做 OCR/Parser 大一统封装。
- 把成功结果归一化成
,再统一写入ParsedFile
。upsert_parsed_files - 结束时完成 pipeline run,并返回单个
风格结果,至少包含:ToolResultdata.library_iddata.run_iddata.file_idsdata.parsed_countdata.failed_countdata.warning_countdata.skipped_countdata.parsed_filesdata.failed_filesdata.warning_filesdata.skipped_files
- 成功或部分成功时,推荐下一步
。build-asset-library
Guardrails
- 不要发明文件内容、OCR 结果、issuer、date 或任何资产字段。
- 不要把目录路径直接当文件列表使用;必须先
。list_local_files - 这是 ingest skill,不负责
抽取、归并、生命周期判断、打包或提交。asset_card - 单文件失败不能阻断成功文件;但 route 决策失败必须明确记 step 与结构化错误。
- 如果 route 决策连续失败,允许 pipeline 保守回退到
,但必须留下 step 和 warning。suggested_route - 低价值或不支持格式应走
,不要硬塞进 parser/OCR。skip - 如果低层提取成功但
失败,返回结构化存储错误,并停止推荐下一步。upsert_parsed_files