Installable agent skills plus reusable Cursor rules for building and debugging LLM agents with tools, streaming UI, and multi-turn sessions.
Skill Claude CodeCodex
Production-hardened practices for building, debugging, and optimizing LLM agents with tools, streaming UI, and multi-turn sessions. Covers dual-store transcripts, artifact ownership, phase gating, orchestration tiers, failure classification, and root-cause fixes over patches. Use when designing agent loops, fixing…
Skill Claude CodeCodex
Preserve agent transcript scroll across app page changes, and give async actions instant press feedback plus in-flight state. Use when building or fixing product shells with a side chat plus main content, when navigation resets conversation position, or when launch/submit clicks feel dead during latency.
Skill Claude CodeCodex
Diagnoses full-stack bugs by inserting console.log at critical state-transition points across the request/response chain, then analyzing where values diverge from expectations. Invoke when a bug is hard to pinpoint through static code reading alone.
Skill Claude CodeCodex
Universal principles for writing agent prompt templates in multi-agent pipelines. Covers role separation, scope steering, anti-patterns, and iterative review flows. Use when designing or refining any LLM agent prompts.
Skill Claude CodeCodex
Align landing hero text and molecule micro-animations into one rhythm for smoother UX. Use when users mention hero animation mismatch, timing drift, overlap, white flash, blank-gap flicker, float-in speed, or visual harmony between text and 3D viewer.
Skill Claude CodeCodex
Diagnose and fix agent tool-call failures in any LLM agent stack: prose promises a tool but nothing runs, toolcalls=0 in logs, wrong tool executes, empty tool payload, streaming UI lag, thinking round-trip loss, or cross-turn transcript amnesia. Use when debugging tooluse / functioncall pipelines with streaming…