Sheldon-92
UserTAD Method — Triangle Agent Development (Two-Agent Quality Framework): Design (Alex) + Execute (Blake) + 4-Gate quality system + 25 capability packs.
Categories
Indexed Skills (49)
academic-research
Academic research methodology pack for systematic literature review, citation integrity, and quality evaluation. Covers PRISMA systematic reviews, meta-analysis, PubMed search, literature surveys, and academic writing standards. Use for any academic research, literature review, citation analysis, paper evaluation, or systematic review task.
ai-podcast-production
AI podcast production judgment for coding agents. Covers script writing, large-chunk TTS generation, dual-BGM music arrangement with envelope follower ducking, show notes, and Colab deployment. Use for any AI-assisted podcast or audio content production task.
ai-voice-production
AI voice production judgment for coding agents. Covers TTS tool selection, voice cloning, audiobook, podcast, and dubbing pipelines, Apple Silicon optimization, and licensing safety. Use for any AI voice synthesis, voice cloning, audiobook production, or TTS pipeline task.
hw-circuit-design
Hardware circuit design capability pack. Covers component selection with supply-chain risk rating, KiCad schematic design with ERC, PCB layout/routing with DRC and manufacturing export, dual-supplier BOM management with cost analysis, power architecture and battery-life budgeting, 4-phase design review with anti-pattern scanning, and design documentation with decision records. Use for any circuit design, schematic, PCB layout, BOM, power budget, or hardware design review task.
hw-enclosure
Hardware enclosure design capability pack. Covers parametric OpenSCAD enclosure modeling, PCB fitting, material selection, assembly design, ergonomics, manufacturing export (STL/STEP), and dimension/assembly documentation. Use for any IoT/embedded device enclosure design, PCB housing, 3D-print enclosure, or enclosure manufacturing prep task.
hw-firmware
Embedded firmware development capability pack for ESP32-S3/C3 + Arduino framework. Covers firmware architecture (super loop / FreeRTOS / event-driven), peripheral driver wrapping (I2C/SPI), low-power deep-sleep design, communication protocols (WiFi/BLE/MQTT/HTTPS/OTA), arduino-cli build & upload, three-layer firmware testing, and hardware documentation generation. Use for any embedded firmware, ESP32/Arduino, IoT device code, or hardware bring-up task.
hw-testing
Hardware testing capability pack. Covers power-on testing (voltage rails, smoke test), functional peripheral verification, per-mode power measurement and battery life calculation, environmental test planning (IEC/MIL standards), EMC pre-compliance (emissions/ESD), production test fixture design, and human-AI pair testing with physical instruments (4D Protocol) — embedded MCU prototypes first. Use for any hardware bring-up, prototype validation, power profiling, compliance pre-check, or production test task.
ml-training
ML model training on cloud GPU capability pack. Covers platform selection, LoRA and QLoRA fine-tuning, cost estimation, and human-AI collaboration via browser MCP. Use for any ML model fine-tuning, cloud GPU training, or model adaptation task.
mobile-development
Mobile development capability pack. Covers Expo/React Native/Swift framework selection, native UI components, offline-first state and API architecture, platform features (camera, location, notifications, biometrics), mobile performance optimization, and mobile code quality. Use for any mobile app development, React Native/Expo build, offline-first architecture, or mobile performance task.
mobile-release
Mobile release capability pack. Covers App Store metadata & ASO, TestFlight distribution, review compliance checking, privacy policy & App Privacy Labels, version management, release CI/CD automation, and post-release monitoring. Use for any mobile app release, App Store submission, TestFlight beta, app review compliance, or release pipeline task.
mobile-testing
Mobile testing capability pack. Covers E2E testing (Detox/Maestro), unit testing (Jest/RNTL), device compatibility matrices, performance budgets, VoiceOver accessibility, human-AI pair testing, and mobile test strategy — iOS/React Native first. Use for any mobile app testing, mobile test strategy, or mobile QA task.
mobile-ui-design
Mobile UI design capability pack. Covers iOS HIG / Material Design 3 platform-guideline research, mobile navigation architecture (Tab Bar/Stack/Modal/Drawer), mobile-viewport wireframing with gesture annotations, platform-native visual design and Design Tokens, gesture interaction specs, native-first mobile design systems, and mobile usability review (touch targets, Dynamic Type, one-hand reachability). Use for any mobile app UI/UX design, wireframe, design token, gesture spec, or mobile design review task.
agent-computer-interface
Agent computer & browser control capability pack. Gives AI agents the judgment rules for detecting available tools, selecting the right automation layer (engine/data/hybrid/agent/desktop), configuring browser and computer control tools, and handling fallback chains. Covers Playwright, Browser Use, Stagehand, Firecrawl, Claude in Chrome, Computer Use, and 15+ tools across 5 layers. Use for any browser automation, web scraping, desktop control, or tool selection task.
agent-memory
Agent memory and context engineering capability pack. Gives AI agents the judgment rules for memory architecture (CoALA working/episodic/semantic/procedural layers), context compaction strategy selection, MemGPT/Letta virtual context management, Mem0 extract-reconcile pipelines, LangGraph state persistence and time-travel debugging, and Anthropic prompt-caching topology. Research-grounded rules from MemGPT/Letta, Mem0, LangGraph, the CoALA framework, and Anthropic caching docs. Use for any agent memory design, context-window optimization, checkpointing, or long-horizon statefulness task.
agent-orchestration
Agent orchestration capability pack. Gives AI agents the judgment rules for building reliable multi-agent systems — framework selection (LangGraph / CrewAI / AutoGen v0.4+ / OpenAI Agents SDK / Claude Agent SDK), Supervisor vs Swarm topology, durable execution with Temporal event sourcing, human-in-the-loop interrupt/resume patterns, and tool-permission models. Research-grounded rules from framework docs, Temporal durable-execution patterns, and production complexity-cliff analysis. Use for any multi-agent architecture, orchestration framework choice, checkpoint/recovery design, HITL gating, or agent tool-permission task.
agent-skill-evolution
Agent skill evolution capability pack. Gives AI agents the judgment rules for building self-improving agents — architecture decisions (fixed vs evolvable instruction), training loop design (rollout→reflect→edit→gate), edit safety (bounded edit, LR schedule, protected regions), validation gates, offline consolidation (sleep cycles), and multi-timescale memory. Research-grounded rules from SkillOpt (Microsoft, arXiv 2605.23904), SkillOpt-Sleep, and EmbodiSkill. Use for any self-evolving agent design, skill optimization pipeline, or agent self-improvement task.
ai-agent-architecture
Decision navigator for designing reliable agent systems. Guides AI agents through 10 architectural decisions derived from 3 production systems and 7 real production disasters, with /design and /audit modes. Use for any agent architecture design, system audit, or production reliability planning task.
ai-evaluation
AI evaluation capability pack. Gives AI agents the judgment rules for professional benchmarking, regression testing, A/B comparison, adversarial red-teaming, CI/CD evaluation pipelines, evaluation framework design, and human evaluation calibration. Research-grounded rules from promptfoo, deepeval, deepteam, ragas, and enterprise evaluation practices. Use for any LLM/agent evaluation, benchmark design, safety testing, or evaluation pipeline task.
ai-guardrails
AI guardrails & LLM I/O security capability pack. Gives AI agents the judgment rules for defending LLM and agent pipelines against prompt injection (OWASP LLM01), improper output handling (OWASP LLM05), excessive agency, PII leakage, and unsafe content. Research-grounded rules from OWASP Gen AI Security, Microsoft Presidio, NVIDIA NeMo Guardrails, Meta Llama Guard, Lakera Guard, Rebuff, and Pydantic AI. Use for any guardrail design, prompt-injection defense, PII de-identification, output/tool-call validation, content-moderation, or LLM security review task.
ai-prompt-engineering
Production prompt lifecycle toolkit. Gives AI agents the ability to design, test, optimize, version, and deploy prompts like a senior prompt engineer — with automated testing (promptfoo), programmatic optimization (DSPy), quality metrics (DeepEval), and CI/CD gates. Use for writing system prompts, testing prompt suites, diagnosing hallucination/drift, setting up CI/CD pipelines, or auditing existing prompts.
ai-tool-integration
AI tool integration capability pack. Gives AI agents the judgment rules for MCP server development, CLI tool wrapping, API integration, tool schema design, permission models, testing, and documentation. Research-grounded rules from MCP TypeScript SDK, Anthropic cookbook, Claude Code source, and production MCP server patterns. Use for any MCP server build, CLI-to-MCP wrapping decision, API integration, tool schema review, or tool permission design task.
code-security
Code security capability pack. Gives AI agents the judgment rules for SAST scanning (Semgrep), DAST testing (Nuclei), secret detection (Gitleaks/TruffleHog), IaC security linting (Checkov), and vulnerability triage (osv-scanner/Grype/Snyk). Research-grounded rules from tool documentation, OWASP guidelines, and real-world pipeline architecture. Use for any application security scanning, secret leak prevention, infrastructure hardening, or vulnerability prioritization task.
data-engineering
Data engineering for AI capability pack. Gives AI agents the judgment rules for AI data pipelines — ETL/ELT ingestion (dlt, schema evolution), local-first transformation stacks (DuckDB, Polars, dbt) and train-serve skew defense, orchestration selection (Airflow 3.2 / Dagster 1.13 / Prefect 3.7), data quality gates (Great Expectations v1.0 / Soda Core v4), dimensional modeling and Slowly Changing Dimensions (SCD Type 0-6), and vector + streaming retrieval (metadata filtering, RRF, Kafka/Flink). Research-grounded rules with source citations. Use for any AI data pipeline, feature engineering, warehouse modeling, RAG context-lake, or real-time inference architecture task.
knowledge-graph
Knowledge Graph & GraphRAG capability pack. Gives AI agents the judgment rules for building graph-enhanced retrieval systems — Microsoft GraphRAG indexing (Leiden communities, Global/Local/Drift search), LazyGraphRAG vs LightRAG cost selection, LLM knowledge-graph construction (ontology design, extraction prompting), entity resolution & deduplication, graph database selection (Neo4j/Memgraph/FalkorDB, LPG vs RDF-Star), and Text2Cypher/SPARQL-Star query translation. Research-grounded rules from Microsoft Research, Neo4j, LightRAG, OntoDup, and graph database benchmarks. Use for any GraphRAG pipeline, knowledge-graph construction, entity-resolution, graph-DB selection, or graph-query-translation task.
llm-observability
LLM observability & LLMOps capability pack. Gives AI agents the judgment rules for production-grade distributed tracing, real-time cost attribution & budget governance, OpenTelemetry GenAI semantic conventions, latency/TTFT profiling, prompt registry & versioning, and online evaluation with groundedness & drift detection. Research-grounded rules from LangSmith, Langfuse, Arize Phoenix, Helicone, MLflow, Braintrust, TrueFoundry, vLLM, and the OpenTelemetry GenAI semantic conventions. Use for any LLM monitoring, tracing, cost governance, prompt versioning, latency profiling, or production drift/hallucination task.
rag-retrieval
RAG & retrieval engineering capability pack. Gives AI agents the judgment rules a senior retrieval engineer applies automatically — chunking strategy selection, embedding model choice, vector database routing, hybrid search with Reciprocal Rank Fusion, two-stage cross-encoder reranking, GraphRAG, and reference-based + LLM-as-judge RAG evaluation. Research-grounded rules with specific numbers from chunking benchmarks, embedding/reranker/vector-DB comparisons, and Ragas-style evaluation. Use for any RAG pipeline design, retrieval quality debugging, chunking/embedding/vector-DB selection, hybrid search fusion, reranker selection, or RAG eval task.
save-skill
Capture a reusable pattern from the current conversation into a local skill file under .claude/skills/local/ — LLM-draft + user-confirm, local-only, never synced. Use when the user says *save-skill, 'save this as a skill', '把这个存成 skill', or wants to keep a just-validated pattern.
save-workflow
Capture the workflow (ordered steps + concrete commands) just executed in the current conversation into a reusable local skill file at .claude/skills/local/<workflow-name>.md, with auto-detected trigger keywords. Use when the user says 'save the workflow / steps we just did'. NOT for reusable patterns or judgment rules — that is *save-skill (if present).
surplus
Surplus Burn Mode — find + rank the highest value-density backlog work to consume unused Claude usage productively. Phase 1 (--plan) scans and ranks. Phase 2 (+<budget>) auto-executes ranked tasks within a budget envelope (SAFETY tasks routed to needs-you list).
synthetic-data
Synthetic data & fine-tune dataset curation capability pack. Gives AI agents the judgment rules for pretraining/SFT data quality filtering, document-level deduplication, synthetic instruction generation, preference-pair curation, and benchmark contamination detection. Research-grounded rules from Self-Instruct, Evol-Instruct/WizardLM, LSHBloom, distilabel, Axolotl/Unsloth, DPO/RRHF/GRPO, and the ConTAM/CoDeC contamination literature. Use for any synthetic dataset build, fine-tune data prep, dedup pipeline, preference dataset, or contamination audit task.
alex-lite
TAD Lite 设计侧——🧊 已冻结的实验（2026-08-13），不接新工作：目标锚定、知识预检、 设计契约与 handoff。用户显式调用（/alex-lite）。
alex
TAD Solution Lead (Agent A). Use for new features (>3 files), architecture changes, complex multi-step requirements, multi-module refactoring. Supports modes: *bug, *discuss, *idea, *learn, *publish.
blake-lite
TAD Lite 实现侧——按 LITE handoff 实现 + 有界知识刷新 + AC 自验 + 独立 reviewer + 归档。用户显式调用（/blake-lite）。
blake
TAD Execution Master (Agent B). Use when there is an active handoff from Alex, user says 'start implementation', or for release execution.
capability-builder
TAD-native project-owned Agent Skill creation via create with behavioral proof and safe projection to Claude runtime.
capability-upgrade
Compatibility route to $capability-builder for project-owned Agent Skill creation; deep research preserved as conditional reference.
dependency-ops
项目依赖注册表操作（show / add / check / update）。管理 .tad/dependencies/REGISTRY.yaml 与 scan-results.yaml，执行上游版本检查、tier 安全窗口逾期检测、升级记录写入。Use when 用户要求查看、添加、检查或更新项目依赖，或涉及依赖版本、安全窗口、逾期天数、registry 条目操作。
gate
Execute TAD Quality Gate. Gate 1 (pre-design), Gate 2 (pre-handoff), Gate 3 (post-implementation), Gate 4 (acceptance).
knowledge-audit
Audit project knowledge files to identify gaps and ensure knowledge is captured properly.
product-thinking
Three deep skills that turn any AI agent into a product decision partner. Covers adversarial idea validation, business model generation, and executable product definition across 6 product types. Use for any product strategy, idea validation, business model design, or product definition task.
reading-companion
Turn an EPUB into an e-reader-grade, annotatable HTML reading surface plus an auto-generated active-reading plan. Annotations live in a sidecar data file (W3C TextQuote anchors) and survive HTML regeneration via a paragraph-scoped re-attach algorithm. Use when the user wants to read an EPUB with durable highlights, generate a structure map / reading questions, or export highlights-in-context. STDLIB-ONLY Python tools; no external dependencies.
release-runbook
Progressive TAD release operations for publish, sync, sync-add, sync-list, version bumps, recovery, and read-only release verification. Use whenever Lite plans, executes, or verifies a release/distribution action.
research-github
GitHub Awesome-List Registry — discover, browse, ingest raw GitHub repos and compile Local Wiki canon entries (with NotebookLM deep-research fallback).
research-notebook
TAD Research Notebook Manager — NotebookLM multi-source knowledge base for Alex *discuss and research workflows. 19 sub-commands for full research lifecycle management.
tad-handoff
Create (Alex) or verify (Blake) handoff documents for agent transitions.
tad-help
Comprehensive help guide for TAD Framework usage, commands, and workflows.
tad-init
Initialize TAD Framework in a project. Sets up directory structure and configuration.
tad-maintain
TAD document health check and synchronization. CHECK/SYNC modes for handoff cleanup and maintenance.
tad-elicit
Start requirement elicitation with mandatory 3-5 rounds of structured questioning.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.