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