trainingsites/campus-ai-os/atlas-setup

atlas-setup — two-mode onboarding

CRITICAL DIRECTIVES

• Conversational only. No forms, no dropdowns, no pre-filled menus. Ask one question at a time and echo the answer back before moving on.
• Tool-agnostic. Auto-detect tools from the live session; never ask the user what tools they have, and never tell them to install one. Zero hardcoded paths.
• Never re-ask anything the host campus onboarding already captured.
• Resolve shared-context filenames through `references/context-files.md` (founding name → canon → alias, each also matched case-insensitively before the next) — never hard-code `shared/icp.md` etc.
• Capture the full identity set (audience, voice, brand, goals, offers) AND, because Atlas is a research team, always ask where to research (sources/communities to prioritize). Write canonical context files, not a self-contained config. Team-private settings go in the ONE namespaced file atlas/config/niche.md (never a bare config/niche.md) — the lowest-precedence fallback, NOT the source of truth.

Step 1 — Detect mode

Glob for workspace.json at the campus root. That is the campus marker, and it is host-neutral — do NOT test for an entry file, whose name differs per platform (PLATFORM.md → ENTRY_FILE), and do NOT assume any particular campus folder name.

Step 2A — founded campus (workspace.json exists) — the normal path

Follow references/campus-mode.md. Resolve and inherit context via references/context-files.md (founding name → canon → alias, each also matched case-insensitively before the next): audience from icp, business model from offers, connections from connections, goals from goals, brand from brand, voice from voice. Never hard-code shared/icp.md etc. Inherit what exists; capture only what's missing. Build the live capability map. Then ask the two Atlas-specific questions a general campus onboarding won't have captured, one at a time: (1) "What do you mostly create — videos, blog posts, webinars/classes, social, workshops, or offers?" and (2) "Where should I research — any specific sources/communities to prioritize (subreddits, competitor channels, podcasts, review sites, industry pubs)?" Save both to the team-private atlas/config/niche.md (output prefs + sources). Show a confirmation summary; allow edits. Do NOT duplicate the inherited shared context — you read it from the resolved location.

Step 2B — campus not founded (no workspace.json)

Do not run the standalone interview. Since v5.1 this skill ships inside the kernel, so founding is /campus-start's job and asking these questions here would interview the owner twice for the same facts. Say: "Your workspace hasn't been set up yet. Run `/campus-start` first to hire your AI team — then I'll pick up the two research questions I still need." Then stop.

<details> <summary>Retired standalone interview — kept for reference, do not run</summary>

Auto-detect tools → report tier ("I can see Firecrawl and a CRM — Premium search, Customer data connected"). First check `./teams-shared/context/` — if canonical files already exist (a peer team set them up), import them and only fill gaps (Spec §5A.5), don't re-ask. One paragraph welcome, wait for "yes". Then ask, one at a time, echoing each:

1. What's your niche / who do you serve? (→ icp)
2. Top 2–3 competitors?
3. Where does your audience hang out?
4. What do you mostly create or sell? (→ business_model + offers)
5. Brand basics — main colors, fonts, visual style? (→ brand; ask because branded outputs use them)
6. Your top goal(s) right now? (→ goals; the council uses these)
7. Where should I research? Any specific places to prioritize or include — subreddits/communities, competitor channels, podcasts, review sites, industry publications? (→ sources in atlas/config/niche.md. Always ask — this is core for a research team. Auto-detected tools set the tier; this sets where to look.)

Also capture voice/tone from how they answer (→ voice). Optional: any specific sources to add (subreddits, podcasts, review pages)?

Then write the canonical shared-context files actually filled (identity, icp, brand, voice, goals, offers) using the §5A.7 field keys in references/context-files.md — only the files captured (partial-capture honesty; never a half-empty file), then hand answers to niche-config-builder.

</details>

Step 3 — Write the team-private config

Hand the two research answers to niche-config-builder, which writes the team-private `atlas/config/niche.md` (Atlas-only settings: business_model, council_voices, search_tier, sources (where to research)). Write this ONE namespaced file — never a bare top-level config/niche.md (that collides with other teams). The campus's canonical context files are the source of truth; `atlas/config/niche.md` is the lowest-precedence fallback. The atlas/config/ directory is scaffolded at founding by campus-start and at upgrade by update-workspace, so it already exists — do not create the campus structure from here.

Step 4 — Registry entry

Since v5.1 Atlas ships inside the kernel, so it is already present in templates/registry.json as a type: kernel entry and needs no self-registration — a kernel team does not register itself the way a third-party team does. If an older type: team entry named strategy-research-team is present from the retired standalone plugin, leave it alone and mention it once; removing another entry is the owner's call, not this skill's. For reference, the registration-block shape a third-party team would use lives in {campus_root}/.campus-os/registration-block-spec.md:

{
  "team": "strategy-research-team",
  "type": "plugin",
  "version": "2.2.2",
  "department": "cross",
  "owns": "Front brain — brainstorming council, niche scans, competitor audits, briefs to execution teams, and outcome reporting; includes grill-me context extraction.",
  "triggers": ["/atlas-think", "/atlas-scan", "/atlas-daily", "/atlas-validate", "/atlas-competitor", "/atlas-grill", "/atlas-report"],
  "done_check": "brief.md + brief.json emitted (front socket)",
  "results": "outputs/{YYYY-MM}/runs/strategy-research-team/{slug}/results.json",
  "permission": "draft-only"
}

Standalone mode: skip silently — there is no campus registry. If the registry file exists but can't be parsed, leave it untouched and tell the owner to run /campus-doctor.

Step 5 — Confirm + first-run nudge

Show the full config summary, allow edits, then nudge: "Try /atlas-think to brainstorm a topic, or /atlas-scan to find opportunities. After an execution team ships, /atlas-report tells you what worked."