Fearvox

Strict parser and validator for Workbench closeout comments, verdict/status discipline, PR reference types, and cross-issue REMAINING sync.

Scout, pilot, or review Mirage virtual-filesystem tool use for Workbench agents while preserving source-of-truth and public-safety boundaries.

Source-first, self-loop resistant guardrails for Capy GitHub dialogue responders before any write-capable PR, issue, or review action.

Disk, swap, VM, agent workspace, and stale session hygiene for Multica workbench runtimes.

Advisory-only algorithm and data-structure review gate between Technical Design and Task List.

Durable memory, session summaries, decision logs, issue closeout synthesis, and compact handoffs.

Use when an operator asks what changed recently, what was left hanging, what can move now, "gm", "where are we", or when a session-derived update needs to become visible to the whole Workbench. Recall recent memory, verify live repo/issue/automation state, surface drift first, and produce a short evidence-labeled action menu.

Use when upgrading a public GitHub repository README, repo metadata, docs map, or first-impression brand surface to match the Zonic/Evensong proof-first standard.

Social platform algorithm consultant — applies Twitter/X's production-grade recommendation patterns to your social data pipeline. Use when building content collectors, recommendation engines, ranking systems, user graph analysis, or content moderation for social platforms (Douyin, TikTok, Xiaohongshu, Instagram, YouTube, etc.). Triggers on: "采集", "collector", "推荐算法", "ranking", "feed", "timeline", "信号", "signal", "graph", "follower", "engagement", "内容分类", "topic", "trend", "热度", "content quality", "spam", "toxicity", or any social-platform data processing.

Load when operator works on a trading strategy, backtest, or live-paper discrepancy and says "回测" / "backtest" / "策略" / "strategy" / "Sharpe" / "Sortino" / "Calmar" / "MFE" / "MAE" / "edge" / "drawdown" / "MaxDD" / "profit factor" / "expectancy" / "winrate" / "confluence" / "regime" / "scalping" / "scalp" / "swing" / "trend follow" / "mean revert" / "breakout" / "market making" / "slippage" / "spread" / "latency" / "tick" / "OHLC" / "Pine script" / "indicator". Also when comparing live paper performance to backtest and the divergence is unexplained. This is the strategy-design counterpart to /fin-deep (which is state query). Use this skill to diagnose **why** a strategy underperforms, recommend **which** pattern to deploy, and ship the actual Pine / Python implementation.

Terminal and HTML visualization for algorithm analysis. Renders complexity comparisons, algorithm profile dashboards, before/after diffs, and full benchmark reports as terminal UI (json-render + ink) or self-contained HTML with Playwright screenshots. Use when data-algo has produced recommendations, profiles, or benchmark results that need visual display. Also trigger when the user says "显示", "show me", "visualize", "画图", "对比", "dashboard", "chart", "compare algorithms", "生成报告", "screenshot", "截图", or wants a shareable report. Trigger proactively after data-algo completes its Ship phase for before/after improvement visuals.

Algorithm and data structure consultant for real codebases. Analyzes code to identify algorithmic bottlenecks, recommends optimal data structures, and implements the chosen approach directly. Use this skill whenever: the user mentions performance issues with data processing, needs a search/sort/filter/graph/cache/queue implementation, asks to optimize loops or nested iterations, mentions Big-O or complexity concerns, needs to choose between data structures, or is building any feature where algorithm selection materially affects correctness or speed — even if they don't explicitly say "algorithm". Also trigger when the user says "太慢了", "optimize", "bottleneck", "数据结构", "排序", "搜索", "缓存", "去重", "遍历", or describes a problem that maps to a classic algorithmic pattern (sliding window, two pointers, BFS/DFS, DP, greedy, etc.).

System design algorithm consultant — applies production-grade distributed systems patterns to your infrastructure. Use when building load balancers, rate limiters, caches, distributed databases, consensus systems, or any feature where system-level algorithm selection affects scalability, availability, or consistency. Also trigger when the user says "system design", "distributed", "load balancing", "rate limiting", "consistent hashing", "sharding", "replication", "consensus", "Bloom filter", "HyperLogLog", "分布式", "限流", "一致性哈希", "分片", "缓存策略", or describes a problem that maps to a classic system design pattern.

Competitive programming template library for ICPC, OI, and Codeforces-level problems. Provides optimized C++ templates for advanced data structures and algorithms. Use this skill whenever: the user mentions competitive programming, ICPC, OI, Codeforces, or needs specific advanced structures like segment tree, fenwick tree, FFT, suffix array, centroid decomposition, heavy-light decomposition, or any contest-level algorithm template. Also trigger on: "线段树", "树状数组", "后缀数组", "快速傅里叶", "点分治", "树链剖分", "competitive programming", "contest", "ICPC", "OI", or describes a problem needing advanced CP techniques.

Perceive before reasoning. Anchor time and retrieve prior episodes before any substantive response. Time receipt + recall as default lookup, not optional luxury.

Classify agent memory into Windburn cognitive-cache objects and route writes safely. Source cache + episodic cache + parking lane.

Wrap remote workhorse runs with prediction → observed-delta → failure-memory packets. Captures the failure cache.

Review source-truth proposals against evidence and boundary rules. Returns verdict — never approves automatically.

Coordinates self-evolution and co-evolution loops within the dash-shatter reverse-engineered CLI. Manages Evensong-style benchmarks as internal reference only, spawns sub-agents via AgentTool, performs vault side-loading from research-vault, and strictly enforces the no-paper-contamination rule. Use when conducting internal evolutionary experiments or preparing observer handoffs.

腾讯会议智能助手，支持会议管理、成员管理、录制与转写查询等功能。⚠️ 使用前必须通过 use_skill 加载本技能，严格遵循当前版本规范。

Bootstrap a new repo with skills/commands/plugins ecosystem, audit its architecture, plan infrastructure from first principles, and verify build stability. Use when onboarding to a new codebase or setting up a project for Claude Code development.

Strudel mini-notation and pattern quick reference. Use when writing or explaining music patterns.

Write Strudel code to a file and trigger the REPL to evaluate it. Use when composing or modifying music patterns for live playback.

Clone Strudel repo, install dependencies, and start the dev server. Use when setting up the project for the first time.

Use when an existing CLAUDE.md in a derived or projected directory (symlink view, build cache, mirror, generated docs, materialized projection) needs to be re-checked against current upstream governance — e.g., upstream rules document was updated, sibling artifact added/removed, manifest schema evolved, or a pre-deploy sanity check is needed on the derived file's guidance.

Use when initializing CLAUDE.md for a derived or projected directory — symlink view, build cache, mirror, generated docs, materialized projection — where the directory's content is regenerated from an upstream source by a tool or script.

Use when an agent needs to turn a public-safe creator idea into a DASH creator capsule, measured artifact, proof route, and remix trail without committing raw generated media or private prompts.

Handle PowerPoint (.pptx) creation, design, and analysis. Use for pitch decks, status updates, and visual storytelling. Use proactively when precise layout positioning and design principles are needed. Examples: - user: "Create a 10-slide deck for the board meeting" -> use design principles + html2pptx - user: "Convert this report into a presentation" -> extract text and map to template - user: "Audit this deck for layout issues" -> generate thumbnail grid for inspection

Multi-agent research vault with Evensong-style memory causation, Ebbinghaus decay, and L2 pressure self-evolution. Three-tier memory (public/restricted/internal), trigger-based agent activation, and 72-hour difficulty-weighted decay algorithm. For research teams building collaborative agent systems with structured long-term memory.

Use when ending a work session — detects project tracking systems, syncs progress, saves session insights to memory, cross-references past learnings, and generates a concise wrap-up report. Triggers on "good night", "wrapping up", "end of day", "signing off", "calling it a day", or equivalent in any language.