Skip to main content

Backfill from Existing Project Notes

Goal​

Seed the knowledge graph from documentation and chat history that already exists, so the graph starts populated rather than empty.

Prerequisites​

  • KMGraph initialized (/kmgraph:init)
  • Existing project notes in one or more of: README, CHANGELOG, existing lessons, Claude/Gemini chat history

Steps​

Option A — Backfill during init​

When running /kmgraph:init, accept the optional backfill prompt:

Would you like to backfill the knowledge graph from existing project context? [y/N]

Select y. The backfill automatically processes:

  • README.md
  • CHANGELOG.md
  • Existing docs/lessons-learned/ entries
  • Existing docs/decisions/ ADRs
  • Chat history exports (if present)

Option B — Backfill after init​

/kmgraph:update-graph --auto --sync-all

This processes all existing lessons silently in one pass without per-lesson prompts. Existing lessons and decisions are never modified — only the search index and graph entries are updated.

Option C — Extract from chat history​

Export chat history from Claude Code or Gemini CLI first:

/kmgraph:extract-chat

The command locates chat logs, extracts lessons and decisions, and presents them for review before writing. Use --delegate knowledge-extractor for large exports (10+ sessions):

/kmgraph:extract-chat --delegate knowledge-extractor

Verify​

/kmgraph:status

The entry count should reflect the backfilled content. Then test search:

/kmgraph:recall "topic from your existing notes"

Next steps​