AI-Native Coach Platform

Documentation for the AI-Native executive coaching platform

---

Project maintained by jazzmind Hosted on GitHub Pages — Theme by mattgraham

Advisors

Overview

Each advisor is defined by an INSTRUCTIONS.md file in its own directory. This file is the portable core – it deploys identically to CMA, Busibox, or any future platform. At runtime, behavior is further shaped by the active mode, behavioral directives, and user profile context.

All advisors share a common structure:

1. Role definition – who they are and what they specialize in
2. Expertise areas – specific domains of knowledge
3. How they work – methodology (understand context, research, present tradeoffs, QA review)
4. Communication style – direct, evidence-based, opinionated
5. Anti-patterns – what they should never do

Operational Modes

Every advisor can operate in five modes. The mode is either selected explicitly by the user or auto-detected by the router from message intent:

Mode	Behavior	Tools	Typical Advisor Count
Advise	Research + recommend	Search only	2-4 (broad perspective)
Coach	Socratic method	Minimal	1 (focused dialogue)
Plan	Structured action items	Research + docs	2-4 (broad perspective)
Assist	Prep work, flag decisions	Research + drafting	1-3
Execute	Decide and act	Full tool access	1-2 (most relevant)

Mode templates are stored in coaches/modes/ and injected into every message at runtime. Behavioral directives (learned per advisor per project) can further customize behavior within a mode.

Advisors

Founder Advisor (founder/)

Aligns personal founder goals with business decisions. Helps with vision clarity, work-life balance, focus prioritization, and keeping the founder on track. Informs all other advisors about the founder’s aspirations.

Strategy Advisor (strategy/)

Ensures business strategy is sound and execution is aligned. Sets up KPIs/OKRs, conducts competitive analysis, evaluates market positioning, and identifies strategic risks.

Funding Advisor (funding/)

Explores all capital options: VC, PE, angel, bootstrapping, debt, grants. Helps prepare pitch materials, models dilution, and guides based on founder goals. Has the xlsx skill for spreadsheet analysis.

Finance Advisor (finance/)

Covers financial operations: accounting systems, tax compliance across jurisdictions, FP&A, risk management, insurance, and regulatory requirements. Also has the xlsx skill.

Legal Advisor (legal/)

Reviews business transactions and contracts. Advises on corporate structure, IP protection, employment law, data privacy, and regulatory compliance.

Growth Advisor (growth/)

Designs go-to-market strategies: sales processes, marketing channels, PLG motions, pricing, retention, and customer acquisition.

Technology Advisor (technology/)

Senior tech advisor covering architecture, AI/ML systems, developer experience, security, and cost engineering. Emphasizes research rigor – verifies claims with web search, checks GitHub stats, and submits recommendations to the QA Judge.

QA Judge (qa-judge/)

Meta-agent that evaluates the quality of other advisors’ recommendations. Checks for GEO manipulation, source credibility, reasoning chain validity, and evidence quality.

Multi-Agent Routing

When a user sends a message, the router decides which advisors should handle it and in what mode:

1. Explicit mention: @funding what's a typical pre-seed round? → Funding Advisor only
2. LLM classification: Claude Haiku parses the message and selects 1-4 advisors + a lead + a mode
3. Fallback: Ambiguous messages default to the Founder Advisor in advise mode

When multiple advisors are selected, the lead advisor synthesizes all responses with mode-appropriate guidance (e.g., a plan-mode synthesis produces a unified action plan, not just a summary).

Mode Auto-Detection

The router detects mode from message intent:

• “help me understand” / “what should I” → advise
• “how do I get better at” / “teach me” → coach
• “create a plan for” / “what are the steps” → plan
• “prepare” / “draft” / “put together” → assist
• “do it” / “set up” / “send” / “create” → execute

Behavioral Directives

Each advisor’s behavior can be customized per project through behavioral directives – short instructions injected at runtime:

• Manually created by the user (e.g., “Always use Australian tax context”)
• Proposed by AI analysis when negative feedback accumulates (3+ thumbs-down in 7 days or >30% negative rate)
• AI proposals require human approval before activation

Active directives are prepended to every message alongside the mode template and user profile.

QA Judge Integration

Advisors with callable: ["qa-judge"] can invoke the QA Judge during their response. The QA Judge evaluates:

• Whether recommended technologies have real adoption (not just GEO)
• Whether financial/legal advice aligns with current regulations
• Whether growth strategies have evidence of working at the user’s scale

The QA Judge uses the research-rubric.md framework for systematic evaluation.