Worktree sessions, elephant memory, PR attribution, and statusline

Auto worktree isolation

A SessionStart hook now auto-creates a worktree immediately before any edits. Every implementation session runs in an isolated git worktree. No more accidental commits to main mid-session.

When you exit plan mode, a companion hook pre-creates the worktree so it is ready before the first file edit.

Human-readable branch names

Worktree branches use whatever you write to .claude/branch-slug instead of a timestamp:

echo "fix-auth-bug" > .claude/branch-slug

Your next edit creates fix-auth-bug instead of impl-20260412-134252-50041.

PR attribution

Every PR created during a Tonone session automatically gets a credit line:

*Atlas · Spine · Warden — [tonone](https://tonone.ai)*

tonone-session-tracker.js records which agents ran; tonone-pr-attribution.js appends the credit block via gh pr edit after gh pr create fires.

Elephant memory

Persistent, caveman-compressed memory that survives session resets.

Two memory files:

• -.elephant/memory.md: local memory for this repo
• -~/.claude/elephant/memory.md: global memory across all repos

elephant-writer auto-captures agent completions, commits, and skill runs. elephant-recall surfaces a startup summary at the top of every session. Run /elephant takeover to cold-start from git history.

Statusline

Tonone installs a rich 3-line statusline:

~/repos/myapp  main  2 dirty  +142 -31 lines
Opus 4.6  ctx: 68%  [===========       ]
5h: 34% on-pace 1.0x  |  7d: 12% slow 0.8x

Line 3 shows five-hour and seven-day token burn rates with a pace multiplier. A 2.0x pace means you are burning twice as fast as the window's linear rate. An optional Line 4 shows your current session goal sourced from .claude/session-goal or .claude/branch-slug.