Menu
Skills in this repository
SkillsMP has collected 184 skills from jluo41/Tools. Open a skill to review its source and details.
jluo41/ToolsShowing 24 of 184 collected skills.
LLM-as-judge evaluator for prediction-interpretation Reports: consumes the report.json from haipipe-individual-inference-report and scores it against a rubric persona via claude_agent_sdk. Each persona defines its own rubric dimensions, scoring rules, and…
Per-individual prediction-interpretation report: loads one individual's data + recent CGM, hits the deployed prediction endpoint, then asks Claude (via claude_agent_sdk) to compose a dual-layer Report -- structured JSON + natural-language text -- for an…
Per-individual endpoint inference test: loads one individual from _WorkSpace/A-User-Store, builds an Endpoint_Set dataframe_records payload, POSTs it to a deployed endpoint (local FastAPI / Databricks / SageMaker -- same wire contract), and prints the…
Per-individual data contract skill: builds/manages a single-individual slice of the pipeline at stages 0-2 (RawDataStore, SourceStore, RecStore) under _WorkSpace/A-User-Store/UserGroup-{dataset}/Subject-{id}. Use to create, inspect, or clean per-individual…
individual-query task-folder specialist: scaffolds {NN}_<name>/ task-folders in the individual task-group (default E-series) that query/visualize ONE individual's data (CGM trace, meal timeline, treatment events) -> results/<run>/{plot.pdf, table.csv}. Called…
model-fitting task-folder specialist: scaffolds {NN}_<name>/ task-folders that fit/train a model (full hyperparam config, real GPU sweep, checkpoint to _WorkSpace/5-ModelInstanceStore/). NOT algorithm dev -- see /haipipe-task-for-algo. Called by /haipipe-task…
evaluation task-folder specialist: scaffolds {NN}_<name>/ task-folders in the eval task-group (default B-series) that score a trained model against an AIData split -> results/<run>/metrics.json. Called by /haipipe-task when task-type=eval. Cross-references…
Unified Stata-engine task-folder specialist: handles all 4 stages internally (cms/case/data/reg), owns the Stata engine contract, the {LNN} stage-letter alphabet, and stage disambiguation. Also a SERVER CHECK mode for the CMS secure server. Called by…
agent task-folder specialist: scaffolds {NN}_<name>/ task-folders in the agent task-group (default F-series) that call an LLM agent with prompts + tools -> results/<run>/{transcript.json, summary.md}. Called by /haipipe-task when task-type=agent. Engine:…
Owns code/haiutils/llm_engine/ -- the unified LLM call runtime for agent task-folders. Two OAuth transports (Claude Agent SDK + Codex OAuth) with API-key fallback. On trigger: CHECK the deployed code exists, CREATE from ref/engine/ if missing, EVALUATE if…
IPO workflow designer + builder + reporter — the basic orchestration unit of haipipe-toolkit: it defines the shared shape (Input → Process[Steps] → Output) and lifecycle (Plan → Build → Execute → Report) that every skill instantiates. Trigger: workflow, plan…
JHU Research-IT helpdesk chatbot. Answers how-to and where-is questions about Johns Hopkins research IT infrastructure by reading the local WIKI/ knowledge base, then giving a short grounded answer with the source page and link. Covers PMAP (Precision…
The clinician-driven inference console — the default inlab-human mode. Patient-first flow in the Claude Code chat: pick a patient from the patient store, see their full chart, list the available prediction models (live Endpoint_Set endpoints), run 'this model…
In-lab human interaction with deployed prediction endpoints — the Claude Code chat is the UI. Tier-1 orchestrator with two modes: CONSOLE (default; patient-first on-demand inference: pick patient → see chart → list models → predict via the endpoint-predict…
Freeze a review_bundle.json for an in-lab reading session: sample de-identified cases, score each via the endpoint-predict tool (live endpoint, deterministic), add narrator-agent explanations, validate schema invariants, freeze. v0.1 source mode: the…
Score an in-lab reading session: join responses.jsonl with the bundle's gold outcomes and compute model accuracy on the sample, clinician blind-vs-assisted accuracy, decision-influence split by model correctness (healthy reliance vs over-reliance — the safety…
Run the in-lab clinician reading session in the Claude Code chat: blind pass (raw case data only → clinician's risk estimate, decision, confidence), reveal (model score + explanation), assisted pass (re-estimate + rate score and explanation SEPARATELY),…
Databricks-specific lessons and knowledge base. 13 hard-won lessons from the REACH-ADHD deployment (VM stockout, init scripts, env vars, pandas compatibility, partition mismatches, MLflow signatures, Model Serving payload routing, CLI auth) plus the full…
Given an audio file, split into sentence chunks, run Azure pronunciation assessment, clip ALL words with timestamps, generate TTS for weak words, call Claude for confusion/Chinese analysis, and output an Obsidian-friendly markdown report with clickable audio.…
Azure + Databricks infrastructure knowledge base. Wiki of core concepts (subscription, IAM, quota, workspace, cluster, Unity Catalog, volumes, notebooks, model serving, user management, cost control), hard-won lessons (gotchas to check BEFORE acting), and an…
Indexes MS365 Outlook email in two modes: monthly-index writes a full-month markdown index, and recent-window summarizes the last N days for chronicle-day-plan. Use for /chronicle-email, chronicle emails, last week's emails, or summarize this week's inbox.
Morning my-task planner. Pulls yesterday's takeaways, overnight emails, and git state as PROMPTS; then walks the user through 3 my-task buckets (paper/project, agent/skill, relationship/career) with sharp Socratic questions; synthesizes today's plan as…
Roll up a day's session .txt files in ~/Daily/<date>/ into one Excalidraw canvas, then mark the day closed and sync to git. Triggered by a 22:00 cron OR on demand. Idempotent — re-running on a closed day is a no-op. After close, new sessions automatically…
Log the current Claude Code session as a concise YYMMDD-HHMM-<slug>.txt under ~/Daily/YYYY-MM-DD/, then sync to the private jluo41/Daily git repo (pull --rebase → commit → push). Use when a working session ends OR on the user's explicit "log this session"…