Tours your users actually finish.
vibe-walk reads your app, names the aha moment, and builds a spotlight tour you own outright — Driver.js module, anchor codemod, six analytics events wired to track what actually lands. And it's honest about fit: if a tour won't help, it tells you before you build one.
Includes9 skills · 3 agents
ValidatedCelestia3 — cycle #16, A/B against the hand-built version.
Find the aha moment, then build the tour.
Two commands. Phase 1 reads your codebase autonomously, names the aha moment, and ranks the stops. Phase 2 runs five interview gates, then builds the tour. (Phase 1 also calls fit honestly — if a tour won't help, you'll know before Phase 2.)
Autonomous surface discovery + verdict.
Reads your orientation docs, route surface, and component composition. Names the aha moment, ranks 8–12 candidate stops, checks anchor-readiness, and delivers the verdict: build / don't build / cheaper first. No questions. Fully autonomous.
Reach for it before you've committed to building a tour.
Five gates, then a complete tour.
Confirms mode (walkthrough vs training), trigger model, substrate, aha moment, and user role — then builds: Driver.js module, anchor-injection codemod, 6-event analytics wiring, and a persistent replay entry point.
Reach for it after discover returns a build verdict.
Three more commands.
Run any directly — no discover required.Bare router. Reads project state, recommends the next step, asks before launching.
Structural self-test — checks plugin files, all SKILLs, scripts, and friction-trigger wiring. Read-only.
L3 self-evolution. Reads session and friction logs, proposes concrete improvements. Never auto-applies.
A tour you own, fully instrumented.
When the verdict is build, Phase 2 emits a drop-in TypeScript module: a spotlightSteps.ts config array and a spotlightTour.ts runner with start(onDone), SSR guard, progress indicator (“3 of 5”), and persistent replay export. Same ownership model as shadcn — you own the generated code from day one.
The anchor-injection pass adds data-tour="<kebab-semantic-name>" attributes via a jscodeshift codemod, auto-injecting only the provably safe subset and routing everything ambiguous to a REVIEW_NEEDED.md with per-item reason codes. Phase 2 halts until you resolve that list. Six analytics events — tour_started, tour_step_viewed, tour_step_advanced, tour_skipped, tour_completed, tour_replayed — wire to Driver.js substrate hooks and emit into your existing analytics system. The success criterion is downstream activation, not completion. Completion is a trap metric.
Steps are capped at 5 (default 3–4). The shape of the completion cliff is consistent across sources and grounded in cognitive-load theory — stated honestly as a conservative guardrail below a single-vendor observed curve, not a proven constant. Dogfooded against the Celestia3 webapp, where the first run caught a real bug the test suite missed.
Two channels.
Stable marketplace
Tagged releases, promoted via the Vibe Plugins marketplace.
/plugin marketplace add estevanhernandez-stack-ed/vibe-plugins /plugin install vibe-walk@vibe-plugins
Canary bleeding edge
Latest main from the solo repo. Fastest path to new substrates and guardrail updates.
/plugin marketplace add estevanhernandez-stack-ed/Vibe-Walk /plugin install vibe-walk
Web only in v1 — the anchor contract requires DOM selectors. Training mode is v2.
One plugin in a family.
Vibe Plugins are a coordinated family — installed independently, composed when present.
