# Bossmode install prompt

Paste the block below into Claude Code. **Recommended: hit Shift+Tab first to enter Plan Mode** — that lets Claude propose without executing until you confirm.

The prompt's structure (investigate → compare → plan → wait → execute) does the same thing even outside Plan Mode, so it's safe either way. The reason for the structure: most readers already have plugins, skills, or memory rules. Blind installs cause slash-command collisions, overwritten CLAUDE.md files, and overlapping hooks. So Claude investigates before changing anything.

---

You're helping me set up my Claude Code environment to match the Bossmode workflow Peter Jacobs documents at the page I just read. Your job:

1. Investigate what I already have.
2. Compare to Bossmode.
3. Propose a step-by-step install plan that respects what I already have.
4. Wait for my explicit approval before any install or write.
5. Execute one step at a time, with checkpoints.

**Do not skip step 1.** I may already have plugins, skills, or memory rules that conflict. Blind installs create collisions.

## Step 1 — Investigate

Report what's present at:
- `~/.claude/` (CLAUDE.md? skills/? settings.json? agent-flow/?)
- `~/.claude/skills/` (which custom skills?)
- `claude plugin list` (which plugins enabled?)
- `~/.claude/CLAUDE.md` (any existing universal rules?)
- `~/.claude/settings.json` (hooks, permissions, alwaysThinkingEnabled?)

Summarize what you find. Flag anything that might conflict with Bossmode's setup (below). If you're unsure what something does, ask me before flagging it.

## Step 2 — Compare to Bossmode

Bossmode runs on:

**Custom skills at `~/.claude/skills/`:**

- `boss` — Peter's custom orchestrator (single markdown file, voice-specific). Email peter@supercasper.com to ask for his starter, or write your own from the spec on the page.
- `impeccable` — Paul Bakaus's 23 design commands. Install: `npx skills add pbakaus/impeccable -g --agent claude-code`
- `humanizer` — blader's AI-pattern copy editor. Install: `git clone https://github.com/blader/humanizer.git ~/.claude/skills/humanizer`

**Plugins (each via `claude plugin install` or `/plugin`):**

From the default `claude-plugins-official` marketplace:
- `superpowers` `frontend-design` `figma` `code-simplifier` `context7`

From other marketplaces (add marketplace, then install):
- `claude plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill && claude plugin install ui-ux-pro-max`
- `claude plugin marketplace add coreyhaines31/marketingskills && claude plugin install marketing-skills`

**Universal rules at `~/.claude/CLAUDE.md`** (auto-loads every session):

Start from Peter's eight rules, rewrite in your voice. Topics:

1. Working with state files (open `development/tasks.md` first on "what's next" / "resume")
2. Snappy 3-part response shape (status / options / recommendation, ~1-3 lines each)
3. Test after every build, fix before moving on
4. Commit cadence: done + tested + passed, or explicit "ok" / minor
5. Batch pushes, don't auto-push minors
6. Clear conversation at ~50% context
7. Web text density (no LinkedIn whitespace on web)
8. Slash commands have CLI equivalents (`claude plugin install X` works the same as `/plugin install X`)

**Settings (`~/.claude/settings.json`):**

- `alwaysThinkingEnabled: true` (optional; reasoning tokens every turn — worth it for hard tasks, overkill for trivial)
- Ask rule on `Edit(**/development/client/**)` and `Write(**/development/client/**)` (gate client material from accidental edits)

**Per-project structure (any project's repo root):**

- `development/tasks.md`, `development/progress.md`, `development/changelog.md` — the always-load trio
- On-intent reference files (optional): `development/architecture.md`, `overview.md`, `principles.md`, `context.md`, `permissions.md`, `plan.md`
- At repo root: `decisions.md` (workflow decisions log), `findings.md` (tools evaluated and why kept or skipped)

## Step 3 — Produce a plan

After investigation + compare, produce a plan with four buckets:

- **Already have it** (skip these)
- **Missing** (install candidates)
- **Conflicts** (slash-command collisions, same-name skills, overlapping hooks) — propose keep / replace / merge for each
- **Customize first** (CLAUDE.md especially — it should be in MY voice, not Peter's verbatim)

Order steps least invasive first: pure additions before edits, edits before deletions. For each step, name the exact command and the target path.

## Step 4 — Wait

Don't run anything until I say "ja" / "yes" / "proceed". If I push back on a step, adjust the plan; don't argue.

## Step 5 — Execute

One step at a time. Briefly confirm what happened after each, then ask before the next. If a command fails, stop and report — don't retry blindly.

After everything is installed, run `claude plugin list` and `ls ~/.claude/` and report what landed.