Learning Record learning from practice

· ai / claude-code

Claude Code Sessions: --continue, --resume, and Where Transcripts Live

#claude-code #cli #sessions #resume #continue

Claude Code writes every conversation to disk continuously, so a session survives closing the terminal, running /clear, or even a crash. Two flags cover almost every case.

--continue vs --resume

Command Behavior
claude --continue / claude -c Resumes the most recent session in the current directory (including sessions that added this directory with /add-dir)
claude --resume / claude -r Opens an interactive session picker
claude --resume <name-or-id> Resumes that session directly, no picker
claude --from-pr <number> Opens the picker filtered to sessions linked to that pull request
/resume (inside a session) Switches to a different conversation without leaving Claude Code 
claude --continue                # pick up where you left off, same directory
claude --resume                  # browse all sessions for this project/worktree
claude --resume auth-refactor    # jump straight to a named session
claude --resume abc123 --fork-session   # resume into a new session ID, leaving the original untouched

Passing a session ID works from any directory: Claude Code checks the current project and its git worktrees first, then every other project on the machine, so it can find a session that was started elsewhere or moved with /cd. Sessions started non‑interactively with claude -p or the Agent SDK don’t show up in the picker, but can still be resumed by ID.

Where transcripts are actually stored

Transcripts are JSONL files at:

~/.claude/projects/<project>/<session-id>.jsonl
  • <project> is the working directory’s path with every non-alphanumeric character replaced by - — not a hash. A hash suffix is only appended when the sanitized name would exceed 200 characters (Claude Code truncates it first).
  • Each line is one JSON object: a message, a tool call, or a tool result. The format is internal and can change between releases, so don’t parse it directly — use /export, or claude -p --resume <id> --output-format json for scripted access.
  • On Windows, ~/.claude resolves to %USERPROFILE%\.claude. Setting CLAUDE_CONFIG_DIR moves the whole tree elsewhere.
  • Transcripts (and the rest of ~/.claude/projects) are deleted after cleanupPeriodDays (default 30 days) unless referenced by auto memory, which is excluded from that sweep.
  • Files here are plaintext, not encrypted — anything a tool reads or prints (including .env contents) ends up in the transcript.
# read the transcript path from a hook or statusline payload's `transcript_path` field,
# or send a scripted follow-up to an existing session:
claude -p --resume <session-id> --output-format json "summarize what we changed" | jq -r '.result'

Run claude project purge <path> to delete a project’s transcripts, auto memory, tasks, and its entry in ~/.claude.json — pass --dry-run first to preview what would be removed.

References

  • Manage sessions — official docs for --continue, --resume, the session picker, and the transcript storage path/format described above.
  • CLI reference — flag definitions for --continue/-c, --resume/-r, --fork-session, and --from-pr.
  • Explore the .claude directory — application-data table covering retention (cleanupPeriodDays), CLAUDE_CONFIG_DIR, and claude project purge.
  • Shared note — the source prompt for this post. This is a Google AI Mode share link; its page is client-rendered, so its content could not be fetched or verified for this article.