/tutorial — Replay the Capability Tour

This is ticket #911 (M3) of the guided-onboarding walking skeleton (technical design: docs/technical-designs/onboarding-increment-1.md, § D3/D4 and the "Standalone /tutorial Reusability Spec"), layered with increment 2's depth adaptivity (ticket #914, § D2/D5/D7) and now grown by #915 (M7) to also render the full teach-in-context glossary — US-6 in full (technical design: docs/technical-designs/onboarding-increment-2.md § D5). /tutorial is a standalone, read-only re-entry point to the same capability tour /onboard shows on first run, plus the plain-language glossary for the five core SDLC terms — for the adopter who skipped either, missed them, or just wants to see them again without faking a fresh fork.

#914 wired /tutorial onto the shared depth-mode marker so it renders in the adopter's current terse/guided mode (one short "why this matters" sentence per section in guided mode; bare, byte-for-byte unchanged rendering in terse — NFR Backward-compat). This ticket (#915) adds the full glossary render after the tour — reading the same shared docs/onboarding/glossary.md asset /onboard's guided-mode asides already read one term at a time (#913) — and lifts the "tour-only" scope note from Rule #4 below. It does not touch the depth-mode marker's derivation, override, or transparency logic (that stays #914's, unchanged), and it does not touch /onboard's asides or the glossary content itself (that stays #913's).

What this skill does NOT do

/tutorial is deliberately disconnected from /onboard, /setup, and /handover — it must work identically whether the fork is fresh, already configured, or predates this feature entirely (US-6 AC). Concretely, it never:

• Reads or writes onboarding.yaml
• Triggers fresh-fork detection (it does not call fresh_fork_state() — see .claude/hooks/_lib-fresh-fork.sh — for gating; that function exists for /onboard and the SessionStart hook, not for this skill)
• Runs any part of the /setup or /handover flows
• Writes to apexyard.projects.yaml or any other registry
• Touches the active-ticket marker or the bootstrap marker

One deliberate, narrow exception (design § D5): this skill MAY both read and write the depth-mode marker (.claude/session/onboarding-depth-mode) — deriving or asking a mode, and honouring a plain-language override mid-/tutorial, is "solicited" teaching (the adopter is running /tutorial on purpose), not a side effect the increment-1 contract forbids. It may also read .claude/session/onboarding-tech-level if present, purely as a derivation input — this skill never writes that file.

Zero OTHER side effects. Running /tutorial should leave git status exactly as it was before the command ran, aside from the gitignored depth-mode marker under .claude/session/.

Process

0. Resolve depth mode (design § D2/D5)

Before rendering, resolve the session's depth mode:

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-onboarding-depth-mode.sh"
mode=$(depth_mode_read)

If a mode is already set this session (from an earlier /onboard run, or an earlier /tutorial invocation), depth_mode_read returns it — terse or guided — and you're done; skip to Step 1.

If no mode marker exists yet (a cold /tutorial on a fork that never ran /onboard), try to derive one from the increment-1 tech-level signal, if present:

signal=$(cat .claude/session/onboarding-tech-level 2>/dev/null || echo "")
mode=$(depth_mode_derive_from_signal "$signal") || mode=""

If $mode is still empty (no usable signal), this is a solicited flow (the adopter ran /tutorial on purpose), so it's fine to ask — use the SAME one-line fallback question increment-1's /onboard D5 uses:

Quick one — have you used git/GitHub before, or is this new to you?

Map the answer (engineer→terse, non-engineer→guided), then write it:

depth_mode_write "$mode"

Never block the tour on this — if the adopter doesn't want to answer, default to terse (the safe default, design § D2/D7) and proceed.

1. Read the shared tour asset

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-portfolio-paths.sh" 2>/dev/null || true

The shared asset path is fixed and framework-relative, not portfolio-relative — resolve it from the git toplevel, not via any portfolio-paths helper:

tour_path="$(git rev-parse --show-toplevel)/docs/onboarding/capability-tour.md"

Read docs/onboarding/capability-tour.md with the Read tool. This is the single source of truth — the same file /onboard Phase 1 reads. Never paraphrase, summarize, or re-word its content; render it close to verbatim (D3 in the technical design — rendering parity between the two consumers is the whole point of the shared-asset seam).

If the file is missing (a corrupted or very old fork), say so plainly and stop:

Can't find the capability tour at docs/onboarding/capability-tour.md — this
fork may be out of date. Try /update to sync with upstream, or check the
file wasn't deleted.

Do not fabricate tour content as a fallback.

2. Render the tour

Print the file's content (the "What's a role?" / "What's a skill?" / "What's a gate?" sections plus the "How the loop works" closer), rendered in the $mode resolved in Step 0:

• terse — render exactly as increment 1 did: no framing, no extra sentences. Byte-for-byte identical to a terse-mode run (NFR Backward-compat).
• guided — after the tour content, add ONE short plain-language "why this matters" sentence (e.g. tying the roles/skills/gates tour back to what the adopter will actually see day to day). Keep it to a single sentence — this is generic tour framing, distinct from the per-term glossary render that follows in Step 3 below.

Open with a one-line frame so the adopter knows this is a replay, not a first-run flow:

Replaying the ApexYard capability tour — the same 60-second orientation
/onboard shows on first run. Nothing on your fork changes by running this.

Then render the tour content.

3. Render the full glossary (design § D5, US-6 full — #915)

Immediately after the tour, read the shared glossary asset:

glossary_path="$(git rev-parse --show-toplevel)/docs/onboarding/glossary.md"

Read docs/onboarding/glossary.md with the Read tool — the same single source of truth /onboard's guided-mode asides (#913) slice one term from. Render all five entries, top to bottom, in file order (issue/ticket, PR, merge, branch, CI). Never re-type, paraphrase, or hand-copy the definitions into this skill file — read the file fresh every invocation, exactly like Step 1's tour read (the increment-1 no-duplication discipline, extended to the glossary asset per design § "Backward Compatibility & Reuse").

A /tutorial invocation is solicited — the adopter ran the command on purpose — so the glossary is always shown, regardless of mode; depth mode only tunes how it's rendered, it never withholds it (this is the same solicited/unsolicited line the design draws for /onboard's asides vs. the on-demand lookup rule, .claude/rules/glossary-lookup.md):

• terse — render the five definitions compactly, one after another, no extra framing.
• guided — render the same five definitions, each with its "Example:" line if the entry has one (most already do — see the glossary's read contract).

Open this section with a one-line frame so it reads as a distinct part of the reply, not a continuation of the tour prose:

And here's the glossary — the plain-language meaning of the five terms
you'll see most often:

If docs/onboarding/glossary.md is missing (same corrupted/very-old-fork case as the tour asset), say so plainly, same shape as Step 1's fallback, and skip straight to Step 4 — a missing glossary should not block the tour render that already succeeded.

4. Depth mode override + transparency (design § D2, FR-6/FR-9)

Same handling as /onboard's — before rendering, and at any later point this session, check the adopter's message for an override phrase or a transparency question:

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-onboarding-depth-mode.sh"
new_mode=$(depth_mode_classify_override "$ADOPTER_MESSAGE")
[ -n "$new_mode" ] && depth_mode_write "$new_mode"

Confirm a switch in one line (Switching to guided — … / Switching to terse — …), same as /onboard. For "what mode am I in?", answer with depth_mode_report — a read, never a write. If the switch happens after the glossary has already been rendered once this invocation, there's no need to re-render it — the switch takes effect on the next /tutorial run or the next /onboard aside, per the Reversibility NFR; /tutorial itself doesn't loop.

5. Close

End with a short pointer back to normal work — no branch, no follow-up questions, no first-win prompt (that's /onboard's job, not this skill's):

That's the tour, and the glossary. Run /tutorial again any time — it's
always safe, always free of side effects.

Rules

1. Never duplicate tour or glossary content. Read docs/onboarding/capability-tour.md and docs/onboarding/glossary.md fresh every invocation; never inline, cache, or hand-copy their prose into this skill file. If either asset's content needs to change, it changes in one place and every consumer (/onboard, /tutorial, the on-demand lookup rule) picks it up automatically.
2. No side effects, except the depth-mode marker. No writes to onboarding.yaml, the registry, the active-ticket marker, or the bootstrap marker — ever. The ONE exception is .claude/session/onboarding-depth-mode (design § D5): deriving, asking, writing, or overriding it is in-contract "solicited" teaching, and it's the only marker this skill may write. Always write it through the shared helper (_lib-onboarding-depth-mode.sh's depth_mode_write) — never a raw echo/redirect.
3. Works on any fork state. Configured, fresh, or pre-dating this feature entirely — the read-and-render behaviour must be identical in all three (aside from depth-mode rendering, which depends on the marker/signal, not fork state). Do not branch on fresh_fork_state().
4. Tour + full glossary + depth-mode rendering (#911, #914, #915). /tutorial now renders both shared assets end to end, respecting depth mode. What it still does NOT do: the per-term, first-encounter guided asides inside /onboard (that stays #913's — a different firing algorithm gated on a separate seen-set marker, § D3) and the ambient any-session single-term lookup (that's .claude/rules/glossary-lookup.md, also #915 — a rule, not this skill, so it fires even when /tutorial isn't running). This skill's contract is still "solicited, whole-asset, side-effect-free render" — it never gates content the way the asides do.

---

Part of ApexYard — multi-project SDLC framework for Claude Code · MIT.