whereamilisted

# whereami "Where are we at / whats next / tldr" is one of the most-repeated prompts in any long-running project, and answers degrade after every compaction. Fix: a per-repo ledger file that outlives sessions, plus a read protocol that checks reality (git/gh) rather than trusting chat memory. ## The ledger: `.claude/state.md` (per repo) ```markdown # State — <repo> Updated: YYYY-MM-DD (session gist) ## In flight - <workstream>: <one-line status> — next: <action> (blocked on: <who/what>, since <date>) ## Done recently - YYYY-MM-DD <thing shipped/merged/decided> ## Sent - YYYY-MM-DD <recipient/channel>: <one-line gist> ## Waiting on - <who> owes <what> since <date> ## Parked / open questions - <item> ``` ## Read mode (default) 1. Read `.claude/state.md` (if missing, offer to bootstrap it from git log + open PRs + this session). 2. **Verify against reality before answering:** `git log --oneline -15`, `gh pr list`, uncommitted changes. If ledger and reality disagree, reality wins — and fix the ledger. 3. Answer as TLDR ELI12: what shipped, what's in flight, what's blocked on whom, the single next action. Short — this is re-orientation, not a report. ## Update mode (`update`, or proactively at the end of a substantial work session) Rewrite the affected sections with what happened this session. Convert relative dates to absolute. Keep the file under ~60 lines — this is a dashboard, not a journal; prune "Done recently" past 2 weeks. ## Why this also fixes parallel-sessi