Imperium Brain

Imperium Brain — Startup Operating System

18 skills, 8 agents, 20 commands. Complete startup intelligence + content creation.

ROUTING TABLE

Read the user prompt. Match keywords to ONE skill (max 2 if cross-domain). Load ONLY the matched skill — never load all skills at once.

STARTUP OPS (9 skills — unchanged)

RESEARCH & INTELLIGENCE (2 skills — new)

CONTENT CREATION (6 skills — new)

Rule: Before generating ANY content, check if brand/brand.json exists. If yes, read brand/brand.json + brand/tone-of-voice.md and apply brand colors, fonts, and voice. If no, use sensible defaults and suggest running /imperium:create-brand.

ROUTING RULES

1. Max 2 skills per request. If a query spans 3+ domains, ask the user to narrow focus.
2. Brand context is lightweight. When loading for content skills, brand.json adds ~150 tokens. Always load it.
3. Multi-skill requests: Follow the PROMPT ANALYSIS PROTOCOL below to detect explicit and implicit multi-skill needs.
4. Startup ops routing is unchanged. All 9 existing skills work exactly as before.
5. Never load skill body until routing is decided. Read the frontmatter/description first.

PROMPT ANALYSIS PROTOCOL

5-step protocol for detecting explicit AND implicit multi-skill needs. Replaces static keyword matching with intelligent intent analysis.

Step 1: Extract Explicit Intents

Scan the user prompt for direct keyword matches against the ROUTING TABLE above.

• Match keywords → candidate skill(s)
• Fast path: If exactly 1 skill matches AND Step 2 finds zero implicit signals → route immediately. No further analysis needed. This handles ~90% of requests.

Step 2: Detect Implicit Intents

Even when the user names only one deliverable, certain language signals reveal hidden prerequisites. Scan for these patterns:

Data signals → research prerequisite needed:

• "data-driven", "backed by data", "with stats", "with numbers"
• "based on market", "based on competitors", "how we stack up"
• "gaps in the market", "whitespace", "underserved"
• "differentiator", "unique angle", "what makes us different"
• "landscape", "who else is doing this", "market overview"

Brand signals → brand prerequisite needed:

• "on-brand", "in our voice", "our tone", "brand-consistent"
• "our colors", "matching our identity", "our style"
• "brand guidelines", "visual identity"

Strategy signals → ceo-advisor prerequisite needed:

• "should we", "what direction", "how to position ourselves"
• "moat", "defensibility", "strategic advantage"
• "where to play", "how to win", "our positioning"

Temporal signals → sequential execution:

• "then", "after that", "based on findings", "once we have", "from the results"

Parallel signals → concurrent execution:

• "and also", "plus create", "across all channels", "simultaneously"

Visual signals → visual-media prerequisite needed:

• "with images", "add photos", "find images for", "include visuals"
• "with screenshots", "product images", "stock photos"
• Content request + carousel plan includes image layouts

Step 3: Check Existing State

Before adding prerequisite skills, check what already exists in the filesystem:

1. Brand check: Does brand/brand.json exist?

   • YES → use existing brand, do NOT add brand-voice as prerequisite
   • NO + brand signals detected → inform user, suggest /imperium:create-brand as prerequisite

2. Research check: Do research output files exist in cwd? (ecosystem-map.md, gap-analysis.md, competitive-*.md, market-*.md)

   • YES → use existing research files as input, do NOT re-run research
   • NO + data signals detected → inform user, add research as prerequisite

3. Neither exists + signals detected: Tell the user what prerequisite is needed and why, then suggest the execution plan.

Step 4: Resolve Dependencies

When multiple skills are identified, resolve execution order using the dependency graph:

Dependencies (must run sequentially):

• research → content (any content skill needs research data first)
• brand-voice → content (brand must exist before content applies it)
• research → ceo-advisor (strategy needs market data)
• research → sales-gtm / marketing (GTM needs market intelligence)

Parallel-safe combinations (same dependency level):

• linkedin + carousel (both consume same research/brand inputs)
• sales-gtm + marketing (both consume same research inputs)
• slides + sop (independent content outputs)

Execution order: Prerequisites first → then primary skill(s), parallel where safe.

Step 5: Confirm if Multi-Step

Choose the right confirmation level based on what was detected:

Quick Reference: Common Implicit Patterns

AGENT COMPOSITION

When to use single vs multi-agent

• Single agent: User request maps cleanly to one domain (e.g., "write a LinkedIn post", "research competitors")
• Multi-agent: User request spans two domains with clear dependency (e.g., "research competitors then create content from findings")
• Max 2 agents per request — same as max 2 skills. If 3+ agents needed, break into sequential requests.

Sequential execution rule

1. First agent runs to completion, saves output to working directory
2. Second agent starts, reads output files from first agent
3. Each agent focuses on its domain — no scope creep across boundaries

Output convention

• Agents write deliverables to the working directory (markdown, JSON, PPTX, etc.)
• File names follow skill conventions (e.g., ecosystem-map.md, gap-analysis.md, top-20-profiles.md)
• Next agent reads these files as input — no need for explicit "handoff" mechanism
• If an agent produces structured data (JSON, tables), downstream agents should parse and reference it directly

Brand as shared state

• Brand is NOT an agent-to-agent handoff — it's shared context available to ALL content agents
• brand/brand.json + brand/tone-of-voice.md are always checked before content creation
• Brand architect creates these files once; all content agents consume them automatically

CROSS-DOMAIN WORKFLOWS

Six common multi-agent workflows. Each specifies trigger, order, handoff, and example prompt.

1. Research → LinkedIn Post

• Trigger: "research" + "post" / "LinkedIn" / "write about findings"
• Order: market-researcher → content-creator (linkedin)
• Handoff: Research outputs ecosystem-map.md, gap-analysis.md, top-20-profiles.md → LinkedIn skill reads findings as source material instead of doing own WebSearch
• Example: "Research the AI coding assistant market then write a LinkedIn post about the top insights"

2. Research → Pitch Deck / Carousel

• Trigger: "research" + "deck" / "carousel" / "presentation" / "slides"
• Order: market-researcher → content-creator (carousel/slides)
• Handoff: Research outputs competitor data, feature matrix, pricing benchmarks → carousel builds positioning deck using real data
• Example: "Analyze our competitors then create a pitch deck showing our positioning"

3. Brand → Any Content

• Trigger: "brand" + any content keyword (post, carousel, slides, SOP)
• Order: brand-architect → content-creator
• Handoff: Brand generates brand/brand.json + brand/tone-of-voice.md → all content skills auto-detect and apply brand colors, fonts, voice
• Example: "Create our brand identity then make a LinkedIn carousel introducing the company"

4. Research → CEO Strategy

• Trigger: "research" + "strategy" / "positioning" / "where we stand"
• Order: market-researcher → ceo-strategist
• Handoff: Research outputs ecosystem map, pricing benchmarks, market gaps → CEO advisor uses for strategic positioning and decision-making
• Example: "Map the competitive landscape for project management tools then help me decide our strategic positioning"

5. Research → GTM Plan

• Trigger: "research" + "GTM" / "go-to-market" / "launch strategy"
• Order: market-researcher → sales-hunter + growth-marketer
• Handoff: Research outputs market map, ICP signals, competitor weaknesses → sales builds outbound strategy, marketing builds inbound strategy
• Example: "Research the developer tools market then build our go-to-market plan"

6. Full Content Pipeline

• Trigger: "brand" + "research" + content keywords (or explicit "full pipeline")
• Order: brand-architect → market-researcher → content-creator (parallel: linkedin + carousel + slides)
• Handoff: Brand first (creates visual/voice system) → research second (creates data/insights) → content last (uses both brand + research as inputs)
• Example: "We're launching next month — create our brand, research the market, then generate launch content across all channels"

7. Visual Media → Content

• Trigger: "find images" + content keyword, OR image layouts in carousel plan with no images provided
• Order: visual-media → content-creator (carousel/linkedin)
• Handoff: Visual-media saves images to media/images/[topic-slug]/ → carousel uses paths in image layouts, linkedin recommends for post
• Example: "Find relevant images about AI coaching and create a carousel with them"

AVAILABLE COMMANDS

Startup Ops (existing)

• /imperium:validate-idea — Startup idea validation
• /imperium:pitch-deck — Pitch deck creation
• /imperium:fundraise-prep — Fundraising preparation
• /imperium:cold-email — Cold email campaigns
• /imperium:competitor-matrix — Competitive matrix
• /imperium:pricing-strategy — Pricing strategy
• /imperium:gtm-plan — Go-to-market plan
• /imperium:metrics-dashboard — SaaS metrics dashboard
• /imperium:founder-kb — Knowledge base search

Research & Intelligence (new)

• /imperium:deep-research — Full 9-phase competitive intelligence
• /imperium:discover-apis — Hidden API discovery for any website
• /imperium:crawl-check — Check imperium-crawl installation & capabilities

Content Creation (new)

• /imperium:create-brand — Brand identity wizard (colors, fonts, voice)
• /imperium:linkedin-post — LinkedIn viral post generator
• /imperium:carousel — PPTX carousel/deck creation
• /imperium:create-video — Video content (bridge to Remotion)
• /imperium:create-slides — HTML presentation slides
• /imperium:create-sop — SOP/runbook/playbook generator
• /imperium:find-images — Image/video sourcing with visual AI review

Eval & Quality

• /imperium:eval-loop — Automated structural checks and quality analysis

AVAILABLE AGENTS

Startup Ops (existing)

• ceo-strategist — Strategic decisions, fundraising, board prep
• cto-architect — Architecture, tech stack, scaling
• growth-marketer — Marketing campaigns, SEO, content
• sales-hunter — Outbound sales, pipeline, GTM
• product-analyst — PRD, roadmap, prioritization

Research & Content (new)

• market-researcher — Deep competitive intelligence, ecosystem mapping
• content-creator — LinkedIn posts, carousels, slides, SOPs, video
• brand-architect — Brand identity creation, voice definition

BRAND SYSTEM

Shared resource at brand/ — generated by /imperium:create-brand.

Files:

• brand/brand.json — 10 colors, 3 fonts, asset paths
• brand/config.json — Output settings (format, naming)
• brand/brand-system.md — Design philosophy, visual rules
• brand/tone-of-voice.md — Voice character, vocabulary, platform adaptations
• brand/templates/ — 5 pre-built voice templates

Usage: Content skills auto-detect and apply brand. No brand = defaults + suggestion to create one.

EXTERNAL TOOL INTEGRATION

imperium-crawl (optional, npm)

Deep research uses imperium-crawl for bulk scraping, AI extraction, YouTube/Reddit mining.

• Check: command -v imperium-crawl
• Install: npm install -g imperium-crawl
• Falls back to WebSearch + WebFetch if not installed.

Remotion (optional, npm)

Video skill bridges to Remotion for MP4/animation creation.

• Check: look for remotion.config.* or package.json with remotion dependency
• Install: npx create-video@latest
• Falls back to HTML slides or carousel if not available.