agent-kernel

Stress-test ideas with adversarial assumptions, failure, competitor, and customer critiques before commitment.

Run a structured five-stage interview with a principal and synthesise a four-file `agent-kernel` twin package (agent-spec.yaml, identity.md, guardrails.md, curated skills, and optional conative profile). Use when a new principal needs a twin from scratch, when an existing twin needs a re-interview to refresh identity or guardrails, or when a partner wants to demonstrate the kernel methodology end-to-end.

Translate a principal's conative profile into authentic twin behavior, tools, guardrails, and evaluation tests.

Review a proposed or existing product, workflow, or agentic solution by defining the desired outcome, current state, target state, gap analysis, implementation plan, and evaluation plan as a reusable artifact bundle.

Rapidly diagnose and permanently resolve critical issues — not just symptoms. Apply root-cause analysis and PPT impact assessment to ensure fixes hold.

Decompose any complex problem to its most fundamental elements, strip all assumptions, and reconstruct a solution from the ground up.

Design a four-tier governance model (Objective → Strategy → Tactic → Action) with explicit reasoning fields at each level, ensuring every automated action is traceable back to a human-authored business objective in under 30 seconds.

Evaluate any idea or initiative using the PPT (People, Process, Technology) framework combined with Revenue, Risk, and Cost dimensions. Produces a scored, comparable assessment with quantitative analysis.

Estimate the probability of success for any proposed initiative or AI deployment based on four factors, identify key risks, and produce a calibrated Go/No-Go recommendation.

Map the downstream consequences of any decision, action, or system change across two causal layers — what the first effect triggers, and what that triggers in turn.

Design a phased go/no-go validation roadmap for R&D or platform teams building systems that will be adopted by multiple downstream teams.

Audit a multi-agent system architecture for gaps in state durability, self-improvement loops, memory/context design, HITL gates, and agent contracts.

Organize any project or initiative into an Immediate / Soon / Later planning matrix, combining Pareto prioritization with OKR-based goal setting.

Break analysis paralysis by identifying the smallest valuable next step, executing it immediately, and building momentum through compounding small wins.

Design systems where behavior is controlled by external configuration files rather than hardcoded logic. Enables non-engineers to manage business rules, enables CI/CD of configuration, and reduces deployment risk.

Apply the athlete's train-and-sprint model to knowledge work. Replace continuous shallow work with deep focus sprints followed by deliberate rest and reassessment.

Create forward momentum for stuck teams by identifying 3 high-visibility quick wins (one each in Revenue, Risk, Cost), executing them visibly, then channeling that momentum into a moonshot vision.

Design coordinated fleets of domain agents (agent factories) at any evolution stage — from a single agent to cross-factory networks — with orchestration, event contracts, HITL governance, and production and digital output coordination.

Evaluate and rank potential AI automation opportunities using a weighted 5-dimension scoring matrix and AI readiness quadrant. Produces a prioritized implementation roadmap.

Design and manage a progressive autonomy system (L0–L3) for AI agents, with defined promotion criteria, demotion triggers, and calibrated review cadences at each level.

Design complete domain-specific AI agents with capability discovery, skill assignments, prompt routing, model selection, action definitions, HITL rules, and cross-agent event contracts.

Apply the Rate of Improvement thesis — measure the metric the business already cares about, track improvement over time, classify the deployment curve shape, and validate whether the AI workflow is succeeding.

Ingest raw business operational data and extract actionable insights framed as automation and improvement opportunities, quantified in business terms.

Produce calibrated confidence scores for any recommendation or action, and design experiment frameworks to measure the actual causal impact of changes before and after execution.

Design Mermaid diagrams to visualize workflows, relationships, governance structures, and system architectures. Select the right diagram type, write compatible syntax, and register diagrams for embedding.

Design human-in-the-loop approval gates and automated guardrails that define the safe operating envelope for autonomous agents. Includes kill switch design for emergency control.

Develop authentic, empathetic leadership by cultivating mindfulness, building unbiased self-awareness, and expressing genuine care through consistent actions.

Give and receive feedback that is both caring and direct. The intersection of caring personally and challenging directly produces the best outcomes for people and organizations.

Design action bundles (tactics) where multiple actions share a logical dependency and must be executed together to achieve the intended outcome. Distinguish bundling necessity from convenience grouping.