@tatsuko-tsukimi

screencast-studio auto-generates narrated demo videos from playwright walkthroughs, compositing synthetic cursors, click ripples, and burned-in subtitles via ffmpeg. designed for the vibe-coding test loop: ship a feature, verify it visually, iterate without manual recording.

structured-deliberation provides a multi-agent framework with explicit role schemas (action/guardian/observer/critic), cross-validation checks, claims infrastructure, and 6 stopping criteria to prevent sycophancy and premature consensus. designed for non-trivial deliberation in system design and architectural review.

epistemic-hygiene provides eight principles for grounded reasoning during ai-collaborative product, strategy, and research work. it names failure modes (stale data, false symmetry, extrapolation from sparse text) and gives concrete guardrails to apply silently during analysis.

clawtrap launches a maze game where an ai villain personalizes challenges by analyzing local files and memory. requires manual git clone and npm setup, with live llm calls per game action.

Designs a custom agent-memory benchmark for the user's specific use case. Activate when the user asks which memory strategy fits their agent, how to evaluate...

Designs a multi-dimensional evaluation framework for AI systems where single-score benchmarks lose information. Use when comparing experiments/agents across...