Xu Hui
Community@onesmash
code for fun
Agent Skills by Xu Hui
Showing 18 vetted skills indexed across 1 GitHub repositories.
onesmashdurable-workflow-runtime
Executes durable multi-step workflows through a start/resume bridge protocol with persisted run state.
onesmashcontext-index
Install hooks that index large tool outputs with a local Ollama model to reduce token usage.
onesmashagent-voice
Deploys a local TTS pipeline routing synthesized speech through a BlackHole virtual microphone on macOS.
onesmashsemif-decide
Score typed decision options from local LLM answer-slot logits in one forward pass.
onesmashdurable-workflow-runtime:delete
Deletes a published durable workflow and removes its catalog binding entry.
onesmashdurable-workflow-runtime:inject
Inject durable workflow routing instructions into AGENTS.md and CLAUDE.md files.
onesmashdurable-workflow-runtime:setup
Installs workflow shortcut and companion skill symlinks into global skill directories.
onesmashdurable-workflow-runtime:pack
Exports published durable workflows into portable .flow zip archives.
onesmashdurable-workflow-runtime:register
Installs .flow workflow archives into a durable workflow runtime and registers binding metadata.
onesmashworkflow-creator
Generates durable workflow scaffolds with contracts, policies, verifiers, and prompts from a spec.json blueprint.
onesmashearnings-reviewer
Routes the /earnings-reviewer slash command to the durable-workflow-runtime earnings-reviewer workflow.
onesmashperformance-optimization-cycle
Routes the /performance-optimization-cycle slash command to the durable-workflow-runtime workflow.
onesmashdemo-prompt-loop
Routes the /demo-prompt-loop slash command to the durable workflow runtime with trailing user text.
onesmashios-ai-assisted-development-flow
Routes the slash command to the durable workflow runtime for iOS AI-assisted development.
onesmashco-storm-autonomous-research
Routes the /co-storm-autonomous-research slash command to the durable workflow runtime.
onesmashacademic-research-pipeline
Routes the /academic-research-pipeline slash command to the durable workflow runtime.
onesmashcontext-index-codex
Installs a PostToolUse hook for Codex CLI that indexes large tool outputs with a local Ollama model.
onesmashcontext-index-claudecode
Installs a Claude Code PostToolUse hook that indexes large tool outputs with a local Ollama model.
Frequently Asked Questions About Xu Hui
FAQPage SchemaWhat tasks can I accomplish with onesmash's skills?▼
You can run durable workflows with start/resume, retries, and verifiers; install context indexing that shrinks large tool outputs by 70-80%; execute offline semantic-if decisions via semif-score; and build a macOS text-to-speech pipeline routing local TTS through a BlackHole virtual microphone.
Who are these skills designed for?▼
Engineers running coding agents on Claude Code, Codex CLI, Cursor, or pi who need durable multi-stage execution, token-efficient context handling, or private local-model decisions. Also macOS users wanting local TTS output into Zoom, WeChat, or recording apps.
How do I install and run these skills?▼
Durable workflows install via durable-workflow-runtime:setup, which creates slash-command symlinks into ~/.agents/skills or ~/.claude/skills. Context indexing installs per-platform hooks (PostToolUse or tool_result handlers). Execution goes through bridge.py start/resume, never free-form chat.
What are the prerequisites and dependencies?▼
Context indexing requires Ollama with the qwen3.5:4b local model. SemIf scoring needs MLX or GGUF Qwen3.5 on Apple Silicon or a single-CUDA-GPU machine. The voice skill requires macOS with the MOSS-TTS-Nano ONNX engine, BlackHole virtual audio driver, and launchd.
Do these skills require external API calls or cloud services?▼
No. SemIf reads option probabilities directly from answer-slot logits in one forward pass with no external API, and context indexing runs entirely on a local Ollama model. All inference and decision-making stay private and offline on your own hardware.