Skills

> **作者**: 邹英龙 > **更新日期**: 2026-04-29（最终版，所有修复已验证） > > 本文档记录Task6开发过程中踩过的所有坑、经验、原则和上下文。仅供开发参考，不属于最终交付内容。 > **适用范围**: Task6 多模态整网变异和验证（单机 + 多机） --- ## 核心原则（不可违背） ### 原则0：CogVideoX / InternVL3 跑通后作为基准 **CogVideoX 和 InternVL3 已成功跑通（单机 + 多机），证明Task6框架、环境配置、日志捕获逻辑、多机部署均正确。** **在调试其他模型时，以下部分禁止修改：** 1. **Task6核心框架**（`utils/task/task6.py`）： - 环境变量读取逻辑（`_init_config`） - PTA/MSA执行流程（`_run_pta_verify`、`_run_msa_verify`） - 多机并发启动逻辑（`ThreadPoolExecutor` + SSH/rsync） - 日志解析正则表达式 - 成功/失败判定逻辑 - 精度/显存/性能判定标准 2. *

>

|

Use when behavior, setup, validation, public wording, data provenance, or workflow commands change and repository documentation may need to be updated.

Use when behavior, setup, validation, public wording, data provenance, or workflow commands change and repository documentation may need to be updated.

Implement and apply an approved change proposal

>

Create a GitHub Task (sub-issue) from the current git diff vs main and link it as a sub-issue of a parent Feature in the local GitHub SmoothAiStockAnalysis. Use when Codex needs to summarize branch changes into a horizontally sliced task with acceptance criteria and create it via `gh`.

Looks harmless but omits important runtime requirements.

>

Master modern business analysis with AI-powered analytics,

>

Canonical task creation skill for this repo. Use when the user invokes initiate-task, asks to start a new project task, classify and approve versioned deliverable contracts, create a tracker-maintenance task with target:tracker, query current task state before creation, coordinate project tasks, or create a primary task with capability_dependencies on usable skills from another known project.

Builds reusable Convex components with isolated tables and app-facing APIs.

Applies safe ESPHome rules for YAML, lambdas, sensors, switches, relays, boot behavior, fail-safes, and validation.

Design and implement a complete ML pipeline for: $ARGUMENTS

Skill file: --- name: ask-questions-if-underspecified description: Clarify requirements before implementing. Do not use automatically, only when invoked explicitly. --- ## Goal Ask the minimum set of clarifying questions needed to avoid wrong work; do not start implementing until the must-have questions are answered (or the user explicitly approves proceeding with stated assumptions). ## Workflow ### 1) Decide whether the request is underspecified Treat a request as underspecified if after explo

**Tier:** POWERFUL **Category:** Engineering **Domain:** Multi-Agent Systems / AI Orchestration --- ## Overview Design production-grade multi-agent workflows with clear pattern choice, handoff contracts, failure handling, and cost/context controls. ## Core Capabilities - Workflow pattern selection for multi-step agent systems - Skeleton config generation for fast workflow bootstrapping - Context and cost discipline across long-running flows - Error recovery and retry strategy scaffolding - Docum

中文漂移审计技能。用于项目或学习过程变乱、上下文漂移、任务分叉、多个方案冲突、命名不一致、Codex 可能顺手改多了时。触发语包括"漂移检查""感觉跑偏了""项目变乱了""检查是否失控""分叉太多""上下文漂移"。

|

Track engagement metrics, measure campaign ROI, and analyze content performance across Instagram, Facebook, YouTube, and TikTok.

Use for writing, reviewing, debugging, or testing Python code that parses, inspects, transforms, builds, qualifies, optimizes, formats, or transpiles SQL with SQLGlot. Trigger on parse_one/parse, sqlglot expressions and scopes, dialect conversion, AST traversal, lineage, qualification, schema-aware rewrites, and unsupported translation. Do not use for executing SQL, database query tuning, regex-only text edits, ORM query construction, SQLFluff lint configuration, or generic SQL unrelated to SQLG

Onboard a new VPS/node to the arifOS federation mesh. SSH key exchange, Tailscale/Headscale setup, organ registration, smoketest, notification

Azure Monitor OpenTelemetry Exporter for Java. Export OpenTelemetry traces, metrics, and logs to Azure Monitor/Application Insights.

Fully autonomous QA testing agent for Modular RAG MCP Server. Reads test cases from QA_TEST_PLAN.md, executes ALL test types automatically without human intervention — CLI commands, Dashboard UI via Streamlit AppTest headless rendering, MCP protocol via subprocess JSON-RPC, provider switches, and data lifecycle checks. Diagnoses failures, applies fixes with up to 3 retry rounds, and records results in QA_TEST_PROGRESS.md. Use when user says 'run QA', 'QA test', 'QA 测试', '执行测试', '跑测试', 'test and

Universal coding standards, best practices, and patterns for TypeScript, JavaScript, React, and Node.js development.

Apply STRIDE methodology to systematically identify threats. Use when analyzing system security, conducting threat modeling sessions, or creating security documentation.

Provides expertise for analyzing DWARF debug files and understanding the DWARF debug format/standard (v3-v5). Triggers when understanding DWARF information, interacting with DWARF files, answering DWARF-related questions, or working with code that parses DWARF data.

Use for Drip Builder work. Builder turns the user's selected Fashion Designer product images into a reviewed, deployed, one-page limited-drop site and writes builder-output.json.

Synchronize reciprocity mandates across agent boundaries.

Use when working with error debugging multi agent review

Embed Photopea in web apps using photopea.js. Covers embedding, file I/O, scripting, exporting, layers, text, filters, and the full Photoshop-compatible API.

Structure a hard business decision — frame the choice, score options against weighted criteria, run a pre-mortem and stress-test, and produce a recommendation + decision record for any active brand

Use when working with context management context restore

Extract properties from a formal spec (TLA+/Lean/ProVerif) into AutoSpec's structured format. Classifies each property with id, statement, essence, and type. Works on formal-spec output or any existing formal spec.

An AI agent specialized in retrieving and answering queries from a laboratory knowledge base, handling documents, regulations, project materials, and supporting image content understanding.

Disciplined diagnosis loop for hard bugs and performance regressions. Reproduce → minimise → hypothesise → instrument → fix → regression-test. Use when user says "diagnose this" / "debug this", reports a bug, says something is broken/throwing/failing, or describes a performance regression.

UMAP dimensionality reduction. Fast nonlinear manifold learning for 2D/3D visualization, clustering preprocessing (HDBSCAN), supervised/parametric UMAP, for high-dimensional data.

>

A Python library providing context managers and patterns for deterministic resource management such as connections, file handles, and cleanup logic in synchronous and asynchronous code.

Run local ComfyUI workflows via the HTTP API. Use when the user asks to run ComfyUI, execute a workflow by file path/name, or supply raw API-format JSON; supports the default workflow bundled in assets.

Design and optimize production-grade multi-agent systems with LangGraph, LangChain, and DeepAgents for complex AI workflows.

An AI agent that assists developers by automatically discovering and injecting project-specific coding guidelines and standards before starting development tasks to ensure code quality and consistency.

A relentless interview to sharpen a plan or design, which also creates docs (ADR's and glossary) as we go.

Implement persistent memory patterns for AI agents using AgentDB. Includes session memory, long-term storage, pattern learning, and context management. Use when building stateful agents, chat systems, or intelligent assistants.

Analyzes multiple provided pages to identify keyword overlap and

Use when a delivery run needs a Stakeholder Input Packet and none exists or the existing one has gaps, OPEN items, hedges, or contradictions — e.g., the project idea lives in a conversation, loose notes, or a short brief. Required before the requirements-analyst agent (and therefore the whole pipeline) can run.

Master Electron desktop app development with secure IPC, contextIsolation, preload scripts, multi-process architecture, electron-builder packaging, code signing, and auto-update.

rosbag（mcap）を使った記録・再生・ループ再生テストと、DDS/QoS 起因の疎通トラブル切り分けの定型手順。「rosbagをループ再生してテスト」「bagを流してもトピックが受信できない」「録画データの周波数がおかしい」「record開始直後のデータが欠けている」ときに使用する。

Analyzes multiple provided pages to identify keyword overlap and