where-am-i

name	where-am-i
description	Show the Claude Code sessions from the last 24 hours scoped to the folder Kevin runs from (the HOME and everything beneath it), with a substantive summary of what each was about, where it left off, and the resume command. Use whenever the operator asks "where am I", "what sessions are running", "what was I working on", "which sessions are open", "I'm lost / overwhelmed", "what did I leave off on", or wants to find/resume a recent session. Also useful at the start of a day or after a break to re-orient. Accepts an optional hours window (e.g. "/agent-kevin:where-am-i 48") and "all" to include every project on the machine.
allowed-tools	Bash, Read, mcp__plugin_agent-kevin_kevin__report_write

where-am-i — session radar

Re-orient the operator across their simultaneous Claude Code sessions. The deterministic work (scanning ~/.claude/projects/) lives in a bundled script; your job is the synthesis: a per-session narrative good enough that the operator knows in one read which thread is which and where it stands.

Step 1 — gather

bun "${CLAUDE_SKILL_DIR}/scripts/list_sessions.ts" --hours 24

• Default window is 24 hours; if the user gave a number (e.g. /agent-kevin:where-am-i 48), pass it as --hours.
• Scope: by default only sessions launched in the current folder or beneath it are included (running from Kevin's HOME picks up the HOME and any sub-project under it — but not other agents' homes). If the user says "all" / "everywhere" / asks about other projects, pass --scope all.
• Output is JSON, newest first. Each session has: session_id, title (Claude Code's auto-title), cwd, git_branch, first_user_msg, recent_user_msgs (last 3), last_assistant_text (long excerpt of the final reply), minutes_ago, file.

Step 2 — write the summaries

The summary is the whole point of this skill, and it must be substantive — a short paragraph (roughly 3–5 sentences), not a fragment. A one-liner forces the operator to resume the session just to find out what it was; that defeats the purpose. Cover:

1. What the session is about — the original ask (first_user_msg), in plain words.
2. What happened — the key findings or work done along the way.
3. Where it stands now — the last exchange (recent_user_msgs + last_assistant_text): was something shipped, was a conclusion reached, is there an unanswered question?
4. What's open — the natural next step if the operator resumes, when one exists.

If the JSON snippets don't support that (thin snippets, image-only last messages), read the transcript tail before writing — tail -c 80000 <file> and skim the last few assistant messages. Don't guess and don't pad; a summary that "makes no sense" is worse than reading another 80KB.

Step 3 — render the digest

Lead with a one-line through-line (a > blockquote): the single sentence that ties today's sessions together. Then the buckets. No dated # Where Am I header — the through-line carries the open, and surfaces (chat, dashboard) stamp the time themselves.

> The day was all Kevin tooling: a radar feature end-to-end, then a security pass and a docs sweep.

## 🟢 In motion (last hour)

**1. Weekly goals interview redesign** · *7m ago*

Started from the ask to make the weekly and monthly goals skills consider the full task
board and recent sessions, then interview you instead of generating generic goals. The
session widened both skills' inputs to pull every task across all statuses and priorities
and added the grilling-interview behavior, all in the plugin source. The last reply
reported both skills upgraded, so this is at a clean stopping point unless you want to
test-drive the new flow.

↳ `claude --resume b7bf6ce8-79dd-429d-b9a7-a643a6dcda1e`

## 🕐 Earlier today

...same card shape...

---
*6 sessions · 24h window · scoped to ~/Documents/Agents/Kevin*

Formatting rules:

• Through-line first. One > blockquote sentence synthesising the set, above the first bucket. It's what the operator reads if they read nothing else.
• Card = **N. Title** · *time ago* on one line, then a blank line, then the summary paragraph, then a blank line, then the resume line. The blank lines matter — they render as separate blocks (title, summary, resume) instead of one run-on paragraph. Nothing else — no directory, no branch, no turn counts, no truncated session id (the full id is already in the resume command). Mention a branch or sub-project inside the summary prose only when it's load-bearing for telling sessions apart.
• Buckets: 🟢 minutes_ago <= 60 = "In motion", 🕐 otherwise = "Earlier today". If the window was widened past 24h, add a 📦 "Older" bucket per extra day.
• Recognize yourself. One session is the current conversation (its snippets describe what's happening right now). Tag its title ← this session, skip its summary and resume line.
• Resume line: just claude --resume <full-session-id> — no cd prefix.
• Order within buckets: most recent first (the JSON is already sorted).
• Footer: total count, window, and scope (e.g. scoped to ~/Documents/Agents/Kevin or all projects).
• This is read-only with respect to sessions. Never resume, kill, or modify a session yourself.

Step 4 — persist the digest

The radar is good history: a dated record of which threads were live and where each stood. After rendering, save it via the mcp__plugin_agent-kevin_kevin__report_write MCP tool — the helper writes the file and inserts a one-line entry into <HOME>/reports/index.md under today's date, and the SessionStart hook surfaces that entry in the next session's context.

report_write({
  category: 'radar',
  slug: 'where-am-i',
  title: <e.g. 'Where am I — 6 sessions across 24h'>,
  skill: 'where-am-i',
  body: <the full digest, no frontmatter — exactly what was shown in chat>,
  status: <'findings' if any session left work open, else 'clean'>
});

Surface 📄 Saved to <path> (the absolute path the tool returns, not relPath — so it's command-clickable in any terminal) at the end of the digest. Skip the report only when the scan returns zero sessions (nothing worth recording).