The system behind the AI work.
I design modular AI systems: agents orchestrate the work, reusable skills provide specialized capabilities, governance keeps actions bounded, and evaluations measure quality. Public pages show the architecture — not the proprietary implementation.
Specialists, not one giant bot.
Each agent has a clear job and pulls in reusable skills as needed. The public portfolio describes the capability and outputs; implementation instructions remain private.
Strategic Operator
Frames ambiguous problems, surfaces constraints, compares options, and turns analysis into decision-ready recommendations.
Meeting → Action
Converts meeting notes into decisions, actions, owners, deadlines, risks, and open questions.
Executive Brief
Compresses complex updates into the signal leaders need: what changed, why it matters, risks, and decisions required.
Project Health
Reviews milestones, dependencies, decisions, ownership, and blockers to identify emerging execution risk.
Process Architect
Finds friction, handoffs, bottlenecks, and automation opportunities, then proposes a cleaner operating design.
Onboarding Agent
Demonstrates role-aware onboarding guidance, task sequencing, and knowledge delivery for new team members.
View public demos →Show the capability. Protect the recipe.
These cards intentionally expose purpose, inputs, and outputs — not the full prompt bodies, routing rules, edge cases, or SKILL.md implementation.
Gut → Data → Gut
Starts with a judgment, tests it against evidence and contradictions, then updates the decision rather than outsourcing thinking to the model.
Risk Radar
Surfaces execution risks from project evidence and distinguishes observed risk from speculation.
Blind Spot
Challenges the current framing and asks what assumptions, stakeholders, constraints, or second-order effects may be missing.
ALT 3
Generates three materially different paths instead of cosmetic variations of the same idea.
X10 Think
Looks for structural leverage instead of incremental optimization: remove, automate, standardize, delegate, or redesign.
Kill Critic
Attempts to invalidate an idea before recommending it, reducing confirmation bias and weak strategy.
PM Mode
Forces a project-manager lens across ownership, sequencing, milestones, dependencies, decisions, and stakeholder alignment.
Executive Signal
Separates important changes, risk, and decisions from informational noise for leadership audiences.
AI Shortcut Library
A broader library of shortcuts and prompt patterns for faster everyday AI work. The complete library lives in the paid product, not this repo.
Get the current library →Shortcuts are not the same thing as skills.
Shortcuts are lightweight command patterns for changing how the model approaches a task. Skills are deeper reusable procedures with logic, guardrails, inputs, outputs, and evaluation criteria. I use both — and keep the distinction intentional.
Think & Analyze
Fast modes for critique, comparison, assumptions, risks, root-cause analysis, fact checking, and alternative viewpoints.
Plan & Decide
Commands for prioritization, decision support, roadmaps, action planning, focus, time allocation, and practical next steps.
Write & Communicate
Rapid transformations for rewriting, tone, clarity, concision, professional messaging, email, headlines, and social content.
Business & Growth
Shortcut modes for offer design, pricing, audience thinking, competitive analysis, market opportunities, content, and funnels.
Learn & Career
Fast modes for teaching, quizzes, study plans, interviews, resumes, mentoring, coaching, and career thinking.
Structure & Automate
Commands for frameworks, matrices, checklists, templates, tables, JSON structures, reverse planning, and automation ideas.
The names can be visible. The implementation stays paid.
The public portfolio shows categories and a few representative commands. The paid library contains the complete command set, original instructions, usage guidance, combinations, and practical examples.
A modular architecture.
The goal is reuse and control: agents do not need to duplicate every instruction if they can route to shared skills and common governance.
AI proposes. Humans stay accountable.
Human-in-the-loop
High-impact actions require human review before anything is sent, assigned, published, or changed.
Evidence before confidence
Recommendations distinguish sourced evidence, inference, and uncertainty instead of presenting everything as fact.
Data minimization
Workflows are designed to use only the information required for the task and avoid exposing sensitive internal data.
Bounded actions
Agents have explicit operating boundaries, escalation paths, and stop conditions rather than unlimited autonomy.
If you cannot measure it, you cannot improve it.
I evaluate agent outputs against dimensions that matter to actual work, not just whether the answer “sounds good.”
Accuracy
Are claims correct and supported by the available evidence?
Completeness
Were important decisions, risks, actions, or constraints missed?
Grounding
Can important claims be traced back to a source or explicit input?
Actionability
Can a person use the output to make a decision or take the next step?
Relevance
Does the output stay focused on the actual job to be done?
Risk & confidence
Are uncertainty, sensitive actions, and possible failure modes represented honestly?