AI Engineering

A router that helps determine which skill or workflow to use for a given task. Use when you need guidance picking the right tool from a suite of AI agent skills.

Craft verifiable goal prompts for AI coding agents like Claude Code, Codex, or Pi. Turns vague tasks into precise goals with checkable completion criteria, ensuring agents run unattended.

Grills the user relentlessly about a plan, decision, or idea to stress-test their thinking. Use when the user wants to explore every aspect or uses any 'grill' trigger phrases.

Generates a concise handoff document from the current conversation, including suggested skills, so another AI agent can seamlessly continue the work. Saves to temporary directory and redacts sensitive data.

Enables an AI agent to teach you a skill or concept through a structured workspace with interactive HTML lessons, reference materials, and learning records, emphasizing long-term retention and practical application.

Turn approved specs and tickets into a verifiable execution goal without re-interviewing the user. Gathers evidence, selects scope, and drafts goals for coding agent sessions.

Performs a five-lens decomposition of project stories to define architecture, record persistent schema decisions, and plan the first slice. Use after repo seeding or when foundational decisions need revisiting.

Use when someone asks what an X2 skill does, which skill to run next, how the gates work, or where an artifact lives. A quick in-flight reference; for the method's rationale, read X2.md at the x2-method repo root.

Guides autonomous implementation of locked software slices, running tests to green with a strict human-turn contract and invariant enforcement.

Diagnoses unknown runtime product misbehavior using structured observation, hypothesis testing, and minimal instrumentation to find root cause efficiently.

Use when starting an X2 project, when the problem domain has no approved epic story set, or when new scope needs stories added to an existing set. Runs problem discovery and produces epic-level stories, then requests gate 1 (stories approved).

Enforce a hard cap and drift-check when a cheap executor consults an expensive advisor, computing effective cost from actual calls instead of assumed rates. Use when adopting the advisor pattern or quoting benchmark discounts.