jnyross

You can run adversarial multi-model code reviews, trace bugs to root causes, measure blast radius of changes, design module architecture before coding, orchestrate parallel worker swarms, write Diátaxis-standard documentation, and maintain auditable decision logs for long-running or unattended engineering work.

Software engineers and technical leads doing code review, debugging, migrations, and architecture design. Skills like typescript-best-practices target anyone editing .ts or .tsx files, while principle skills guide developers wiring validation, designing types, or sequencing refactors and multi-step changes.

Skills like interrogate spawn multiple LLM reviewers that challenge changes from independent angles, while blast-radius executes real code to prove safety. The show-me-your-work skill keeps a TSV decision log with what, why, evidence, and result per decision for human review.

Skills are invoked by name or trigger phrases within an agent environment. The why skill discovers available MCPs to query source control, issue trackers, docs, chat, observability, and error tracking. The setup-pstack skill configures which models pstack uses per role based on detected available models.

Yes. The swarm skill fans out N parallel workers and returns one consolidated report, arena spawns N competing candidates and grafts the strongest parts together, and figure-it-out designs auditable playbooks for large migrations with hypothesis loops and logged decisions.