Skip to main content

curt-convert

Convert Claude Code session transcripts (.jsonl) into .curt single-file conversation files. Sister tool to curt-wildchat.

Install​

npm install -g curt-convert   # or: npx curt-convert

CLI​

curt-convert ~/.claude/projects/C--tmp-demo/abc-123.jsonl
curt-convert ~/.claude/projects/C--tmp-demo/ -o ./out
curt-convert "**/*.jsonl" -o ./out
curt-convert session.jsonl --stdout > session.curt
  • A .jsonl file, a directory (walked recursively), or a glob all work.
  • Output defaults to alongside each input; -o/--out redirects.
  • Names follow the bulk convention via curtFilename().
  • Bulk runs never overwrite: colliding inputs (parent sessions and sidechain files share a sessionId) get a deterministic -<6 hex> suffix from the source path — stable across runs. Set CURT_CONVERT_SELFCHECK=0 to skip the self-check.

Mapping highlights​

  • sessionId → conversation.id; ai-title (or first user text, ≤60 chars) → conversation.title.
  • tool_use blocks → toolCalls[] and tool_result blocks → separate role: "tool" messages with toolCallId — declares the tool-linkage extension.
  • Thinking blocks kept out of content as a message extra; the opaque signature dropped.
  • Documented losses: token counts and cost, parentUuid chaining, <system-reminder> harness spans, meta/noise lines.

Handling warnings​

Output is transcript-equivalent, not redacted: if the transcript contains secrets, the .curt file contains them too. Scrub before sharing.

Where to go deeper​

The full mapping table (every JSONL line type, kept or dropped), the documented-losses ledger, and the security note live in the repo README: github.com/JordanNewell/curt-convert.