A Claude Code skill that turns Claude into a live, interactive tour guide for any codebase — a hands-on tutor that walks you through how the code actually works, one digestible concept at a time, pausing at real checkpoints so you can ask questions, go deeper, or change direction at any moment.
This is the conversational counterpart to walkthrough-generator skills: instead of handing you a static document or HTML file, the tour is the session.
- Grounds every claim in real code. Every explanation cites
file:linereferences from files Claude actually read — no invented snippets, no fuzzy paraphrasing. When unsure, it goes and looks. - Hands you the wheel — genuinely. Each stop ends with a real checkpoint offering concrete next directions. Your questions aren't interruptions to the plan; they become the plan, and the tour resumes when you're ready.
- Calibrates depth as it goes. Starts at competent-engineer-new-to-this-repo altitude, then adjusts from your questions — unpacking fundamentals on request, speeding up when you're ahead of it.
- Draws when a picture beats prose. ASCII sketches, Mermaid, or richer diagram tools depending on what your client renders.
- Scales to large codebases by refusing to "cover" them: it fans out parallel explore agents to map the terrain, then tours the ~20% that carries the mental model. A 200K-line codebase is a focused sitting, not a slog.
Copy the skill into your global skills directory:
git clone --depth 1 https://github.com/jonascript/guided-tour /tmp/guided-tour && mkdir -p ~/.claude/skills && cp -R /tmp/guided-tour/skills/guided-tour ~/.claude/skills/Or install with skills:
npx skills add https://github.com/jonascript/guided-tour --skill guided-tourNew skills are picked up at session start — open a fresh Claude Code session after installing.
Invoke it directly:
/guided-tour path/to/repo
Or just ask naturally in any repo:
- "give me a guided tour of this codebase"
- "walk me through how authentication works — and let me ask questions"
- "I'm new to this project, where do I start?"
- "onboard me to this repo"
The tour proposes a short itinerary, gets your nod, and starts. Interrupt it whenever you like — that's the point.
A condensed moment from a real session (touring the Redis source):
Claude: …so
serverCronis Redis's janitor — expiry sweeps, hash-table resizing, replication pings all hang off this one heartbeat, re-armed by its own return value (return 1000/server.hz;— server.c:1562).Checkpoint:
- Stop 2 — follow real bytes: a client sends
GET fooand we trace it toprocessCommandat server.c:4455- Linger — where the loop's timers are stored (it's charmingly unsophisticated)
- Or take us somewhere else
User: ok so if I somehow turned off the janitor and there were no scheduled tasks, it would sleep forever?
Claude: Exactly right — and we already saw the line that proves it. At ae.c:380:
struct timeval tv, *tvp = NULL; /* NULL means infinite wait. */— no timers → the kernel is told "no deadline at all." One precision though: you turned off the alarm clock, not the doorbell. A client socket still wakes it instantly. And here's the fun part — you can now predict what breaks: keys you touch still expire (lazy path), but snapshots never trigger, idle clients never time out, and my favorite:SIGTERMis ignored — the signal handler only sets a flag that serverCron reads. You disabled its ability to die politely.
Questions redirect the tour; the tour follows, then offers the way back.
- Articulate your model back. Explaining a stop in your own words and asking Claude to grade it point-by-point is the fastest way to find the gaps.
- Chase tangents. "Could this pattern apply to my project at work?" produces some of the best sessions — the skill follows you there and back.
- Propose designs. Try "what if it worked like X instead?" and let the tour review your idea against the real code's constraints.
MIT