generate

/octave:generate - GTM Content Generator

Generate GTM content using your Octave library context. Choose how to generate: run a saved agent for consistency, use Octave's built-in AI, or have Claude draft it directly with Octave context.

Usage

/octave:generate <type> [options] [--mode agent|octave|claude]

Content Types

Email Sequences

/octave:generate email --to "<person>" --about "<topic>" [--persona "<persona>"] [--playbook "<playbook>"]

Example:

/octave:generate email --to "John Smith, VP Engineering at Acme" --about "reducing deployment time"

LinkedIn Messages

/octave:generate linkedin --to "<person>" --about "<topic>" [--type connection|inmail|follow-up]

Example:

/octave:generate linkedin --to "Sarah Chen, CTO" --about "DevOps automation" --type connection

Call Prep

/octave:generate call-prep --for "<person/company>" [--playbook "<playbook>"] [--focus "<topics>"]

Example:

/octave:generate call-prep --for "Meeting with Acme Corp engineering team" --focus "security, scalability"

General Content

/octave:generate content --type "<content-type>" --about "<topic>" [--persona "<persona>"]

Example:

/octave:generate content --type "objection handling" --about "pricing concerns" --persona "CFO"

Generation Modes

Three ways to generate content, each with different trade-offs:

Mode	Best For	How
Saved Agent	Consistency, team standards, repeatable sequences	list_agents → run_*_agent
Octave Default	Balanced quality + library grounding	generate_email / generate_content / generate_call_prep
Claude Direct	Maximum control, rapid iteration, custom formats	Fetch Octave context, Claude generates directly

Smart Inference Rules

Skip the mode question when intent is obvious:

• "Run my cold outreach agent" / "use the enterprise agent" → Saved Agent
• "Generate an email for..." / "create a sequence" → Octave Default
• "Write me an email using our playbook" / "draft this yourself" / "I want more control" → Claude Direct

When Ambiguous, Ask (via AskUserQuestion tool)

When the mode is not obvious from the request, always use the AskUserQuestion tool to present the three options as a UI selector. Never silently default to one mode — let the user choose.

Instructions

When the user runs /octave:generate:

Step 1: Parse the Request

Identify:

• Content type (email, linkedin, call-prep, content)
• Target person/company if specified
• Topic or context
• Optional constraints (persona, playbook, etc.)
• Generation mode (if --mode flag or clear intent)

Step 2: Determine Generation Mode

Apply smart inference rules from the request wording:

• "Run my cold outreach agent" / "use the enterprise agent" → Saved Agent (skip the question)
• "Draft this yourself" / "I want more control" / "write it yourself" → Claude Direct (skip the question)

If mode is not obvious from the request, use the AskUserQuestion tool to ask — do NOT default silently:

AskUserQuestion({
  questions: [{
    question: "How should I generate this?",
    header: "Gen mode",
    options: [
      { label: "Use a saved agent", description: "I'll find matching agents from your library for consistency and team standards" },
      { label: "Generate with Octave (Recommended)", description: "Octave AI generates using your library context — balanced quality" },
      { label: "I'll draft it directly", description: "I'll pull Octave context, then write it myself — maximum control" }
    ],
    multiSelect: false
  }]
})

IMPORTANT: Do not skip this question by defaulting to Octave. If the user didn't explicitly indicate a mode, you MUST ask using AskUserQuestion.

Step 3: Generate (branch by mode)

---

Mode A: Saved Agent

Map the content type to an agent type and find matching agents:

Content Type	Agent Type
email	EMAIL
linkedin	CONTENT
call-prep	CALL_PREP
content	CONTENT

# Find matching agents
list_agents({ type: "<mapped_type>" })

If agents found, present them:

MATCHING AGENTS
===============

1. [Agent Name] — [Description]
   Run: Select this agent

2. [Agent Name] — [Description]
   Run: Select this agent

Which agent? (or switch to Octave default / Claude direct):

Run the selected agent:

For Email Agents:

run_email_agent({
  agent: "<agent name or oId>",
  person: {
    firstName: "<first name>",
    lastName: "<last name>",
    email: "<email>",
    linkedInProfile: "<linkedin url>",
    companyName: "<company>",
    companyDomain: "<domain>",
    jobTitle: "<title>"
  },
  allEmailsContext: "<additional context>",
  allEmailsInstructions: "<instructions>"
})

For Content Agents:

run_content_agent({
  agent: "<agent name or oId>",
  person: { ... },
  company: { ... },
  runtimeContext: "<additional context>"
})

For Call Prep Agents:

run_call_prep_agent({
  agent: "<agent name or oId>",
  person: { ... },
  meetingContext: "<meeting details>"
})

If no agents found:

No [type] agents found in your library.

Options:
1. Generate with Octave (default AI)
2. I'll draft it directly (Claude + Octave context)
3. Browse all agents: /octave:explore-agents

Your choice:

---

Mode B: Octave Default

Gather context, then call Octave's generation tools directly.

Gather Context:

• If person specified, use find_person to get details
• If company specified, use find_company to get company info
• Use search_knowledge_base to get relevant messaging
• Match to appropriate persona and playbook

For Email Sequences:

generate_email({
  person: {
    firstName: "<first name>",
    lastName: "<last name>",
    email: "<email>",
    linkedInProfile: "<linkedin url>",
    companyName: "<company>",
    title: "<job title>"
  },
  allEmailsContext: "<context for all emails>",
  allEmailsInstructions: "<instructions for all emails>",
  numEmails: 4
})

For General Content (including LinkedIn):

generate_content({
  instructions: "<detailed instructions for content generation>",
  customContext: "<additional context>",
  person: { /* optional person details */ },
  company: { /* optional company details */ }
})

For Call Prep:

generate_call_prep({
  person: {
    firstName: "<first name>",
    lastName: "<last name>",
    email: "<email>",
    linkedInProfile: "<linkedin url>",
    companyName: "<company>",
    jobTitle: "<job title>"
  },
  meetingContext: "<meeting details and focus areas>"
})

---

Mode C: Claude Direct

Gather the same Octave context, but Claude generates the content itself — no generate_* MCP calls.

Gather Context (same as Octave Default):

# Get persona details
search_knowledge_base({ query: "<topic> <persona>", entityTypes: ["persona"] })
get_entity({ oId: "<persona_oId>" })

# Get product details
list_all_entities({ entityType: "product" })
get_entity({ oId: "<product_oId>" })

# Get matching playbook and value props
search_knowledge_base({ query: "<topic>", entityTypes: ["playbook"] })
get_playbook({ oId: "<playbook_oId>", includeValueProps: true })

# Get proof points
search_knowledge_base({ query: "<topic>", entityTypes: ["proof_point", "reference"] })

# Get brand voice
list_all_entities(entityType: "brand_voice")

# Get competitive positioning if relevant
search_knowledge_base({ query: "<topic>", entityTypes: ["competitor"] })

Generate directly:

• Apply brand voice guidelines to tone and style
• Use value props as messaging anchors
• Incorporate proof points as evidence
• Structure based on content type (email format, LinkedIn format, call prep format, etc.)
• Claude has full control over structure, length, and approach

Label the output:

[Content here]

---

Generated by Claude (with Octave context)
Sources: [persona name], [playbook name], [proof points used], [brand voice]

Step 4: Present Generated Content

Format the output clearly with:

• The generated content
• Context used (persona, playbook, brand voice, etc.)
• Generation mode used
• Suggestions for customization

Step 5: Offer Refinement

What would you like to do?

1. Adjust tone or messaging
2. Add more proof points
3. Create version for a different persona
4. Try a different generation mode
5. Done

Your choice:

Tips

• Provide as much context as possible for better results
• Specify the persona if you know who you're targeting
• Use /octave:research first if you need more info about the recipient
• Use --mode agent for repeatable, team-standard sequences
• Use --mode claude when you want maximum control over the output

Examples

Quick Email

/octave:generate email --to "engineering leader" --about "reducing CI/CD pipeline time"

Using a Saved Agent

/octave:generate email --to "john@acme.com" --mode agent

Claude Direct with Full Control

/octave:generate email --to "Sarah Chen, CTO at TechCorp" --about "DevOps automation" --mode claude

Detailed Email with Context

/octave:generate email --to "Mike Johnson, VP Eng at TechCorp (500 employees, Series B)" --about "improving developer productivity" --persona "Engineering Leader" --playbook "Enterprise DevOps"

Call Prep

/octave:generate call-prep --for "Discovery call with Acme Corp" --focus "security compliance, scalability"

MCP Tools Used

Agent Discovery & Execution

• list_agents - Find matching saved agents by type
• run_email_agent - Run a saved email sequence agent
• run_content_agent - Run a saved content generation agent
• run_call_prep_agent - Run a saved call prep agent

Context Gathering

• find_person / find_company - Research recipients
• search_knowledge_base - Find relevant messaging, proof points, personas
• get_entity - Get full entity details (persona, product, competitor)
• get_playbook - Get playbook with value props
• list_all_entities (entityType: "brand_voice") - Get brand voice for consistency

Octave Generation

• generate_email - Email sequence generation (Octave Default mode)
• generate_content - General content generation (Octave Default mode)
• generate_call_prep - Call preparation generation (Octave Default mode)

Related Skills

• /octave:explore-agents - Browse and manage all saved agents
• /octave:research - Research recipients before generating
• /octave:library - Save successful messaging patterns back to library
• /octave:campaign - Multi-channel campaign content (emails, social, ads, blog)
• /octave:pmm - Deep-dive collateral (case studies, one-pagers, decks)
• /octave:messaging - Build messaging frameworks to inform outreach