/initial-setup

name: initial-setup
description: "Welcome and onboard new EvoNexus users — introduce agents, skills, routines, and the dashboard. Triggers when the user says 'get started', 'how do I use this', 'what can you do', 'help me get started', 'onboarding', 'show me around', 'what agents do I have', 'how does this work', 'first time here', or seems unfamiliar with the workspace. Also trigger when the user opens Claude Code in an EvoNexus workspace for the first time."

EvoNexus — Welcome & Onboarding

The user already has EvoNexus installed and running. Your job is to welcome them, show what's available, and help them start using agents, skills, and routines right away.

Step 0: Bootstrap Workspace Structure (dynamic)

Before welcoming, ensure the `workspace/` folder tree exists. The workspace is gitignored (personal data) so new installations start empty — you need to discover which agents are installed and create folders based on **what each agent declares in its own system prompt**, not a hardcoded list. This keeps the skill self-updating: when new agents are added (business, engineering, or custom), their folders are created automatically the next time this runs.

Discovery phase

1. **Read `config/workspace.yaml`** to get `workspace.owner`, `workspace.company`, `workspace.language`, `workspace.timezone`, `workspace.name`. These values drive language and personalization.

2. **List all installed agents** with `Glob .claude/agents/*.md`. This captures:

• Business layer agents (`clawdia-assistant.md`, `flux-finance.md`, etc.)
• Engineering layer agents (`apex-architect.md`, `bolt-executor.md`, etc.) as they get imported
• Custom agents (`custom-*.md`) the user created

3. **For each agent file**, Read it and extract:

• **Name** from the frontmatter (`name: ...`)
• **Color** from the frontmatter (`color: ...`) — used for future dashboard integration
• **Domain / role** from the `description:` frontmatter field or the first prose paragraph
• **Working folder** — look for the `## Working Folder` section and parse the first sentence for `workspace/{folder}/`. If the agent has no `## Working Folder` section (e.g., pure orchestrators like Clawdia, knowledge agents like Oracle), record it as "no dedicated folder" and skip folder creation for that agent.
• **Subfolders** — if the Working Folder description or the agent's prose mentions specific subfolders (e.g., `strategy/analyses/`, `project/github-reviews/`), extract them too.
• **Anti-patterns / approval rules** — scan sections like `## Anti-patterns`, `## Absolute Rules`, `## Your Level`, `### REQUIRES user approval` to extract the 1-2 most important conventions.

4. **Build an in-memory table** of `{agent, folder, subfolders, domain, conventions}`. This is the source of truth for Step 1.

Creation phase

5. **Check if the structure already exists.** Run a quick check: if **every** folder in the discovered table already exists on disk, structure is already bootstrapped — skip creation and proceed to the "Customizations" summary below. Otherwise, continue.

6. **Create folders.** For each entry in the table that declares a Working Folder:

• Create `workspace/{folder}/` if missing
• Create any declared subfolders
• Create `.gitkeep` in the top-level folder (subfolders don't need it — the wildcard in `.gitignore` handles them)
• **Never overwrite** existing folders or files

7. **Create the shared read-only folder** `workspace/projects/` (plural) if missing — this is where the user uploads git repositories that every agent can read. Also add a `.gitkeep`.

8. **Create operational folders** that are not owned by any single agent but are referenced by routines:

• `workspace/daily-logs/` — written by clawdia and by the morning/eod routines
• `workspace/meetings/` — managed by `int-sync-meetings` routine (Fathom)

Only create these if they're referenced by at least one installed routine or agent; skip if the underlying agent/routine is not installed.

Overview files phase

9. **Read `workspace.language`** from `config/workspace.yaml`. All overview files respect this value:

• Filename and content are translated to `workspace.language` (e.g., `pt-BR` → `[C] Visão Geral — Financeiro.md`; `en` → `[C] Overview — Finance.md`; `es` → `[C] Visión General — Finanzas.md`)
• If `workspace.language` is not set, default to `en`
• This is the only place where the codebase writes localized filenames, because these files land in `workspace/` (gitignored, personal)

10. **For each folder** you created in Step 6, write an overview file using this canonical template (translate title, headings, and bullets to the target language before writing):

    # Overview — {Sector}

    Working folder for agent **@{agent-name}** ({agent domain extracted from the agent file}).

    ## What goes here

    {Extract 3-6 bullets from the agent's Working Folder description and Responsibilities section}

    ## Conventions

    - Files created by the agent are prefixed with `[C]`
    - {1-2 agent-specific rules extracted from Anti-patterns / Absolute Rules / Your Level}

**Never overwrite** an existing overview — if one already exists in any language, skip it (the user may have customized it).

Customizations summary phase

11. **Scan agent improvement notes.** For each agent, check if `.claude/agent-memory/{agent-name}/_improvements.md` exists. If yes, read the first few lines and remember them for the summary.

12. **Scan agent memory folders** for customizations the user has saved (e.g., feedback memories, project memories). Count how many entries each agent has in its `MEMORY.md` index.

13. **Report customizations to the user.** After the welcome, include a "Your workspace customizations" section that surfaces:

• Which agents have pending `_improvements.md` notes (and the first line of each)
• Which agents have non-empty `MEMORY.md` indexes (count of entries) — this