Problem or use case
Entire only captures Agent session data/ any session data when hooks fire during a live session. Users with months of pre-existing transcripts under ~/.claude/projects//*.jsonl have no way to bring that history into Entire — entire session list, entire checkpoint explain, and token-usage aggregates all start from zero on adoption day. Current Issue only captures a probable solution for claude code though.
The main issue is the cold start of a repo onto entire and if I already have the transcripts available locally for my Repo, I should be able to add it to the context atleast, to use dispatch, recap and other entire features but not have first class features like rewind, checkpoint etc.
Desired behavior
Add a new top-level command:
entire import claude-code [--path DIR] [--dry-run] [--since DATE] [--session ID]... [--force]
By default it auto-scopes to ~/.claude/projects/<SanitizePathForClaude(repoRoot)>/*.jsonl (the standard Claude Code layout for the current repo) and idempotently writes each transcript into Entire’s checkpoint metadata.
Proposed solution
Imported checkpoints are read-only history. They preserve:
per-turn transcript slices (one checkpoint per user-prompt turn, matching live-session granularity)
token usage (input / output / cache create / cache read, with streaming dedup by message.id)
subagent token rollups (via the existing agent-.jsonl lookup)
prompts, model name, timestamps, original cwd
They are not rewindable. Live sessions snapshot working-tree state at every step; historical transcripts don’t have that, and reconstructing it from Write/Edit tool_use blocks is fragile (misses out-of-band edits, conflicts with later commits). Faking a snapshot would be dishonest. entire checkpoint rewind should refuse with a clear message.
Alternatives or workarounds
No response
Problem or use case
Entire only captures Agent session data/ any session data when hooks fire during a live session. Users with months of pre-existing transcripts under ~/.claude/projects//*.jsonl have no way to bring that history into Entire — entire session list, entire checkpoint explain, and token-usage aggregates all start from zero on adoption day. Current Issue only captures a probable solution for claude code though.
The main issue is the cold start of a repo onto entire and if I already have the transcripts available locally for my Repo, I should be able to add it to the context atleast, to use dispatch, recap and other entire features but not have first class features like rewind, checkpoint etc.
Desired behavior
Proposed solution
Imported checkpoints are read-only history. They preserve:
per-turn transcript slices (one checkpoint per user-prompt turn, matching live-session granularity)
token usage (input / output / cache create / cache read, with streaming dedup by message.id)
subagent token rollups (via the existing agent-.jsonl lookup)
prompts, model name, timestamps, original cwd
They are not rewindable. Live sessions snapshot working-tree state at every step; historical transcripts don’t have that, and reconstructing it from Write/Edit tool_use blocks is fragile (misses out-of-band edits, conflicts with later commits). Faking a snapshot would be dishonest. entire checkpoint rewind should refuse with a clear message.
Alternatives or workarounds
No response