skills

Use when setting up a new project for structured AI-agent development - scaffolds a knowledge-base harness with architecture docs, workflow phases, ADRs, patterns, and verification types with worktree isolation

Interactive, multi-phase GitHub PR review with human-in-the-loop triage before posting. Use when asked to "do a thorough PR review", "interactive review of PR", "review PR with triage", "comprehensive review of PR #N", or any request for a full/deep/structured code review of a PR. Also triggers on "review PR and post comments" or "walk through PR findings with me". This skill runs specialized review agents in parallel, deduplicates findings across categories, presents them to you in configurable batches so you decide what to include, then posts the final review as inline GitHub comments — ensuring every finding that goes up has your explicit approval.

Autonomous optimization loop — measures metrics, scores pages by opportunity, runs experiments, and makes data-driven keep/discard decisions. Use when running the optimization cycle, checking experiment status, or after deploying changes to evaluate impact. Use PROACTIVELY when the user mentions optimization, A/B testing, experiments, metrics review, or site performance.

Use when asked to explain, understand, walk through, or summarize a GitHub pull request. Triggers on phrases like "explain this PR", "walk me through PR

Use when defining product requirements, creating feature specifications, or establishing contracts between stakeholders - creates verification-agnostic acceptance criteria in Gherkin format with multi-modal verification support and contradiction analysis