kb-ingest

KB Ingest — Document Intake + Classification + Linking

Process new documents into the Knowledge Base. Reads content, classifies into the right PARA location, generates metadata, creates wikilinks, updates related documents, and maintains indexes.

Inbox/ is optional, not a prerequisite. Apply this workflow when the user directly asks to create, add, organize, or substantially integrate knowledge in the wiki. A direct write in its final PARA folder still performs ANALYZE, metadata, linking, index, validation, human log, and build telemetry steps; only MOVE is skipped.

Usage

/kb-ingest                      # Process all files in Inbox/
/kb-ingest path/to/file.md      # Process a specific file
/kb-ingest path/to/folder/      # Process all .md files in a folder
"위키에 이 내용을 만들어줘"       # Direct creation; route through the same workflow

Execution Flow

For each document to ingest:

0. START BUILD TELEMETRY

When telemetry is enabled, start one explicit ingest operation before reading or classifying source material. For a batch, start one operation for the whole batch, not one per file. Retain the exact operation_id and parent request_id printed by the bundled helper; reuse both in the summary.

PARA_KB_PLUGIN_ROOT="${CLAUDE_PLUGIN_ROOT:-${PLUGIN_ROOT:-}}"
python3 "$PARA_KB_PLUGIN_ROOT/scripts/para_kb_telemetry.py" build-start <<'JSON'
{"source":"kb-ingest-skill","source_kind":"inbox|direct|batch"}
JSON

Set source_kind from the actual entry route: inbox for an Inbox-origin document, direct for a direct wiki request or specified file, and batch for a multi-document run. The same bundled path works in Claude Code and Codex. Telemetry failure must not block a valid ingest, but do not invent or recover an operation id from a later unrelated request.

1. READ

• Read full document content
• Extract existing frontmatter (preserve if present)
• Identify document type: paper, note, reference, meeting, idea, log, etc.
• Note content length (affects summarization decision)