菜单
Standalone text cleaning for entity-linker pipelines. Detects source type and applies the right cleaner (ASR, email, markdown, HTML, plain). TRIGGERS: "clean up transcript", "parse speaker turns", "normalize timestamps", "strip email headers", "flatten markdown", "preprocess this text".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | input-preprocessing |
| description | Standalone text cleaning for entity-linker pipelines. Detects source type and applies the right cleaner (ASR, email, markdown, HTML, plain). TRIGGERS: "clean up transcript", "parse speaker turns", "normalize timestamps", "strip email headers", "flatten markdown", "preprocess this text". |
| model | inherit |
| tools | ["Read"] |
Reusable text preprocessing for any pipeline that ingests text. The entity-linker skill calls this automatically; other pipelines (Echelon, intelligence agent) can also use it standalone.
from entity_db.preprocess import clean, detect_source_type
source_type = detect_source_type(raw_text) # or pass --type flag
clean_text = clean(raw_text, source_type)
| Type | Detection | Cleaner |
|---|---|---|
email | From: <email> header | Strip headers, signatures, quoted replies |
asr | [HH:MM:SS] timestamps | Strip timestamps, speaker labels, fillers |
markdown | # heading patterns | Strip formatting, preserve text |
html | <html> tag | Extract text, strip tags and entities |
plain | fallback | Pass-through (no-op) |
From:, To:, Cc:, Subject:, Date:, etc. at top.-- followed by name block).> lines, On DATE, wrote: blocks).references/asr.md — ASR-specific noise patternsreferences/email.md — email header list, signature heuristicsreferences/markdown.md — markdown element strippingreferences/html.md — HTML entity handlingreferences/plain.md — plain text conventionsCatalog CRUD and staging review with Apps-first UX and elicitation fallback. TRIGGERS: "review staged entities", "approve new entities", "add entity to catalog", "rename entity", "deprecate entity", "import seed", "manage catalog".
End-to-end entity resolution: reads a text file or pasted content, runs preprocessing, calls resolve_link_text, handles disambiguation via the Disambiguation App (or elicitation fallback), and writes the annotated output. TRIGGERS: "link entities in this", "tag mentions", "normalize names in call", "resolve entities against catalog", "link-file", "link-text".
Reference skill for the matching pipeline: scoring model, normalization chain, Czech inflection rules, phonetic algorithms, and thresholds. TRIGGERS: "tune matching", "why did it miss this name", "explain scoring", "add a phonetic algorithm", "adjust thresholds", "debug matching".