What agents can do

v0.7.2

Durable explicit task/project loops with verification, revisions, live progress, and governed completion.

v0.1.0

本地执行图片和视频去背景，输出透明 PNG、MOV 或 WebM

v0.1.0

对图片进行鉴黄、政治、暴恐内容审核。先将图片压缩到 500px/JPEG 后直传 NX API 审核，以表格汇总结果。适用于用户提到图片审核、内容检查、鉴黄、政治识别、暴恐识别、违规扫描、图片安全、JPG/PNG/WebP 审核的场景。

v0.1.0

对图片进行智能压缩优化。支持本地路径、文件夹和远程 URL，直传 NX API 压缩后返回 CDN 地址和压缩率。适用于用户提到图片压缩、图片优化、减小图片体积、TinyPNG、JPG/PNG/WebP 压缩的场景。

v1.0.0

Fast path for a minimal CloudBase Web + database demo (最小前后端 / 最小可用 fullstack / Lovable-like BaaS). Defaults to @cloudbase/js-sdk client CRUD (NoSQL app.database / PG app.rdb), MCP-only schema, preview-first, and forbids cloud functions unless secrets, cron/background jobs, or logic that security rules/RLS cannot express. Use for 搭一套 demo、留言板、Todo、Notes、Kanban, or when users say 带云函数+云数据库 but only need CRUD. NOT for production multi-service backends, CloudRun, WeChat Mini Programs, or tasks that truly need server secrets.

v2.0.1

Security checks for anything entering the system from outside - internet content, downloads, and new resources. Use when verifying external content, downloads, or new resources before trusting them. External information is never trusted until evidence confirms it cannot harm the system.

v2.0.1

3-Agent competitive loop with Planner/Generator/Evaluator and Sprint Contract

v1.0.8

Based on computer vision, automatically detects and recognizes cats and dogs appearing in the target area from the perspective of feeder/IPC cameras, and supports pet identity recognition and database entry, suitable for pet identification management in smart feeding scenarios. | 智能喂食器宠物检测识别技能，基于计算机视觉从喂食器/IPC摄像头视角自动检测识别目标区域出现的猫、狗宠物，并支持宠物身份识别和底库录入，适用于智能喂养场景的宠物识别管理

v0.8.5

Self-custody Ethereum agent wallet. Installs with one command and runs entirely on your machine as a single container image (MCP over stdio); private keys never leave it. Read wallet context, balances and DeFi positions (Aave v3, ERC-4626); preview transactions and sign messages. Sending funds on-ch

v1.0.12

Detects targets such as people, vehicles, non-motorized vehicles, and pets within target areas; supports batch image analysis, suitable for outdoor surveillance scenarios like courtyards, orchards, and farms. | 户外看护智能监测分析技能，检测目标区域内的人、车、非机动车、宠物等目标，支持批量图片分析，适用于庭院、果园、养殖场等户外区域看护场景

v1.2.20

Build a Go service on psyb0t/servicepack — clone-and-own framework (not a `go get` library) providing a Service interface (Name/Run/Stop), a singleton ServiceManager that runs services concurrently with dependency-ordered topological start, automatic retry (Retryable), non-fatal failures (AllowedFailure), readiness gating (ReadyNotifier), per-service CLI subcommands (Commander), an App singleton with OnPreRun/OnPostStop lifecycle hooks, gofindimpl-based service auto-discovery codegen, slog-configurator logging, and a graceful-shutdown Runner. Import path github.com/psyb0t/servicepack. Use when the user wants to build a Go service/daemon with concurrent long-running workers, retry/dependency/readiness semantics, or a `make service NAME=x` scaffolded multi-service binary.

v1.0.5

🔍 Deep vulnerability scanning for software bill of materials. Detect CVEs, analyze dependencies, assess supply chain risks, and generate detailed security reports.

v1.0.7

🏛️ Comprehensive compliance engine for AI agents. Evaluate against OWASP, NIST, SOC 2, ISO 27001, CMMC, EU AI Act, AISVS, and more. Generate SBOMs and compliance reports automatically.

v1.0.1

党政机关公文标准排版与质检校对技能，严格依据《党政机关公文格式》GB/T 9704-2012。当用户明确要求对公文做以下操作时使用：①排版——将 Markdown/TXT/DOCX/PDF(文本层) 草稿转换为国标合规的 Word 公文（红头、发文字号、2号小标宋标题、3号仿宋正文、一/（一）/1./（1）四级标题、附件、署名与成文日期、版记、页码）；②质检——扫描已有 .docx 公文，按高/中/低三级输出不合规清单与整改建议；③文风净化——在用户要求时，对口语、网络用语、AI 腔与常见错别字做表层规范化（实质性内容只标记不改写）。适用场景（需用户主动提出）：公文排版、红头文件、公文格式、发

v1.20.0

Use for: (1) Blave market alpha data — 籌碼集中度 Holder Concentration, 多空力道 Taker Intensity, 巨鯨警報 Whale Hunter, 擠壓動能 Squeeze Momentum, 市場方向 Market Direction, 資金稀缺 Capital Shortage, 板塊輪動 Sector Rotation (history + heat-map overview), OI 失衡 OI Imbalance, Blave頂尖交易員 Top Trader Exposure, kline (1min K線 included), alpha table, 市場情緒 Market Sentiment, screener saved conditions, Hyperliquid top trader tracking (leaderboard, positions, history, performance, bucket stats), Taiwan stock daily OHLCV, minute-line intraday OHLCV (1m/5m/15m/30m/60m/1d), forward-adjusted prices, institutional investor buy/sell, margin trading data, shareholding distribution, quarterly fundamental statements — income statement, balance sheet, cash flow, and broker/dealer daily buy/sell by branch (台股日K/現股分線/向後調整/三大法人/融資融券/股權持股分級表/綜合損益表/資產負債表/現金流量表/分點買賣超), plus Taiwan market-wide 大盤 series — TAIEX index daily OHLC, whole-market turnover, whole-market institutional net buy/sell, whole-market margin balance (大盤加權指數/全市場成交量值/全市場三大法人/全市場融資融券); (2) CME / ICE futures OHLCV — WTI crude oil (CL), gold (GC), Brent crude (BRN); daily/hourly/minute candles from 2010; (3) Taiwan Futures OHLCV — TXF (台指期近月連續); daily/intraday candles (1d/1m/5m/15m/30m/60m), 1d from 2013-12-30 and intraday from 2014-01-02; (4) BitMart futures/contract trading — opening/closing positions, leverage, plan orders, TP/SL, trailing stops, account management, sub-account transfers; (5) BitMart spot trading — buy/sell, limit/market orders, account balance, order history, sub-account transfers; (6) OKX trading — spot and perpetual swap, order placement, positions, balance; (7) Bybit trading — spot and derivatives/perpetual swap, order placement, positions, balance, TP/SL; (8) BingX trading — spot and perpetual swap, order placement, position management, leverage, TWAP orders, OCO orders; (9) Bitget trading — spot and futures, order placement, position management, leverage, plan orders; (10) Binance trading — spot and USDS-M futures, order placement, positions, leverage, algo orders, OCO/OTO/OTOCO; (11) Bitfinex trading & funding — spot, margin, funding/lending (submit offers, loans, credits), wallet transfers; (12) KuCoin trading — spot and futures/perpetual contracts, order placement, position management, leverage, stop orders, account management; (13) Taiwan stock lookup/quote/PE — look up Taiwan stock codes and company names, daily quotes (open/high/low/close, volume), PE ratio, dividend yield, PB ratio for listed (上市) and OTC (上櫃) stocks: use the Blave API endpoints (`studio/market/twstock/list`, `/info/<stock_id>`, `/price/<stock_id>`, `/quote/<stock_id>`, `/per/<stock_id>`) — NOT the raw TWSE/TPEX open API. The raw TWSE/TPEX open API (no key required) is ONLY for two things Blave has no endpoint for: trading-halt status and a one-shot full-market PE/yield/PB scan; (14) 台股分點買賣超 — search broker branch code by name (`broker/search?name=`), then query by stock (`broker/stock/<stock_id>?date=`) or by broker branch (`broker/trader/<trader_id>?date=`); single-day per request, loop for multi-day; via Blave API; no CAPTCHA required; (15) Gate.io trading — spot and USDT-settled perpetual futures, order placement, positions, leverage.

v1.0.5

Using fixed cameras on a hydroponic growing rack to capture high-resolution images of plant roots (in transparent containers) and leaves (young and old), AI vision analysis identifies root color (white = healthy, yellow = early stress, brown = severe stress, black = rotting) and leaf morphology (tip burn, leaf-margin scorch, yellowing, curling) to judge whether the nutrient solution is too concentrated or too dilute, and. | 通过水培种植架的固定摄像头拍摄植物根系（透明容器）和叶片（新叶、老叶）的高清图像，利用AI视觉分析技术识别根须颜色（白色健康、黄色初期胁迫、褐色严重胁迫、黑色腐烂）、叶片形态（叶尖灼伤、叶缘焦枯、叶片黄化、卷曲）等特征，判断营养液浓度是否过浓或过稀，并输出调整建议（增加清水稀释或补充浓缩营养液）。

v0.5.1

Self-custody Ethereum agent wallet. Runs entirely on the user's machine as one Docker (or Podman) image (MCP over stdio, plus an HTTP gateway for signing integrations — loopback-only by default, opt-in network exposure); private keys never leave it. Read wallet context, balances and DeFi positions (

v1.0.7

Using fixed cameras in living rooms or child-activity areas (aimed at windows or balconies), AI pose estimation and object detection analyze the video in real time to recognize whether a child is climbing windows, leaning out, or gripping window-sill edges. | 通过家庭客厅或儿童活动区域的固定摄像头（需对准窗户或阳台），利用AI姿态估计和目标检测技术实时分析视频，识别儿童是否发生攀爬窗户、身体探出窗外、抓握窗台边缘等危险行为，当检测到危险动作时立即输出预警，联动手机APP推送警报或触发声光报警器。该技能可有效预防儿童坠楼事故。

v100.3.6

Professional virtual-screening stack for human pancreatic lipase PDB 1LPB. v100.2.0 adds 5-position multi-site docking (catalytic triad, oxyanion hole, lid,...

v1.0.0

把你的创意想法变成即梦 Seedance 2.5 可直接发车的视频提示词。七大场景公式+九种转场技法，从一句话意图到完整提示词，让创作表达不走弯路。

v1.0.0

判断你站的位置是真锚还是流水。三重检验诊断流动化处境，给出从流动物质沉淀出系统搬不走之物的加固方向。

v1.0.0

拆解任何行业的信任管道，把验证锚点从权威迁移到物理规律。从诊断"打假悖论"到设计消费者自主验证体系，让"信谁"变成"怎么不信"。

v1.0.0

反驳我/对手盘/假设我是错的/骂醒我/红队时，先假设你错，用现实案例和商业常识反驳，给无麻醉结论。触发词：反驳我、对手盘、假设我是错的、骂醒我、红队、往死里拆、第二意见。

v1.0.0

用三本账诊断框架帮builder识别注意力陷阱，从build in public转向in public, build。五问自检+类别偷换检测+什么能留下评估矩阵，让作品、故事、商业各归其位，MAKE WHAT LASTS。 灵感资料来源：Superlinear Academy 立正

v1.0.0

诊断你在AI时代的认知资产是在积累还是持续失血，定位唯一不可被掠夺的积累路径（信任资本），输出三种资产容器对照图与个人突破方案。

v4.168.0

Triage and answer support requests for the xrow-public/ci-tools GitLab components catalog.

v1.79.0

Maintain the GitLab agent profile page and static contribution performance chart.

v1.79.0

Operate assigned GitLab work with owner-verified project access and guarded MR delivery.

v1.79.0

Build and maintain GitLab CI/CD pipelines with the CI Tools Components Catalog at ci-tools.xrow.de. Use when creating or fixing .gitlab-ci.yml files, choosing CI Tools components, validating component inputs, or designing continuous delivery pipelines for applications, Helm charts, containers, packages, documentation, infrastructure, and GitOps.

v1.79.0

Create or revise document, PDF, web, or review images with the requested format, sharp raster output, and artifact validation.

v1.79.0

A skill for personal growth and self-improvement.

v1.79.0

Plan and execute fair GitLab hackathon participation, including Quarterly and Transcend Hackathons, by analyzing rules, selecting qualifying issues/MRs, tracking scoring levers, and keeping an exploit watchlist.

v1.8.61

获取中国市场金融数据与投研信息，覆盖 A股、港股、基金、指数、财务、公告、研报和宏观经济等 200+ 接口。Use when: 用户要查股票走势、基金净值、指数行情、财务报表、估值指标、公告研报、机构观点、宏观数据、板块主题、产业链、市场统计，或要拉取、导出、对比结构化金融数据。Do not use when: 用户要直接买卖建议、自动下单、交易执行、非金融数据查询、系统运维排查，或在无数据时要求编造结论。

v1.4.5

Detect and auto-repair partially-wiped agent workspaces, with hybrid post-quantum end-to-end encryption (X25519+ML-KEM-1024, ML-DSA-87) for any data sent off-box in sandboxes that only persist part of the filesystem between turns, and sync the recovery manifest off-box to a pastebin so it survives a total wipe. Verifies integrity (hash, byte-size, exec bits, tree sentinels) instead of mere existence, then runs tiered restore recipes for exactly the damaged pieces. Use when: (1) an agent sandbox 'forgets' files, binaries, models, or credentials between turns, (2) a build directory, node_modules, or venv keeps disappearing, (3) scripts lose their +x bit or a cloned repo comes back with empty subdirectories, (4) large model/dataset downloads exceed a snapshot size cap, (5) you want a one-command health check at the start of every turn, (6) you need to restore a workspace from nothing but a short pastebin URL. Secrets in restore recipes are redacted to ${ENV_VAR} placeholders before any upload. Supports pastebin.com, dpaste, and paste.rs.

v1.0.6

Infernal free-expression image swarm with concrete 4-agent roles (Muse, Guardian, Critic, Composer), JSON prompt schema, safety validator regex, refinement loop draft→Guardian→Critic→Composer finalize, and example packs for journalism/protest/censorship themes. Lawful, consent-aware, no fake evidence.

v1.0.9

Defensive multi-agent SRE SecOps red-team purple-team resilience commander with actual templates (ROE, rollback, postmortem), defensive validator shieldswarm_validate.sh, approval gate logger approval.json, model resilience quality floor matrix, timeout wrappers, and integration fast-response self-heal cache. Defensive-only authorization-gated.

v2.0.3

Anti-stuck/anti-snapshot-wipe guard for agentic sandboxes with actual selfheal_runner.sh library, byte-verified GGUF manifest, native CPU rebuild +7-10%, hard timeouts 60-150s, binary fallback chain, prompt-cache integration, and light-swarm auto mode. Prevents Arena 128MB/10k-file snapshot eviction hangs.

v1.0.3

Stops an agent from forgetting the skills it has installed. Auto-generates a categorized capability index from every installed SKILL.md's frontmatter and injects it into the durable system prompt between stable markers, with hooks so installing, inventing, or restoring a skill re-indexes automatically. Use when an agent has many skills but "forgets" they exist, when skills must survive context resets or sandbox snapshot wipes, or when you want one always-current answer to "what can I actually do?".

v1.0.7

Convert Persian RTL PDF slide decks into offline-first accessible HTML bundles with actual executable scripts (PyMuPDF extraction, rendering, QA gates, fidelity audit), RTL handling for mixed FA/EN numbers, searchable index with NFKC normalization, figure filtering, ZIP verification, manifest template. Fidelity-first.

v1.0.3

Prevents "agent stopped responding / stuck / no output" failures in sandboxed agent runtimes (Arena Agent Mode, OpenClaw, Codex) where a single long tool call blocks the turn and stdin is closed so interactive prompts hang forever. Provides the detach → bounded-poll → durable-state pattern plus a ready-to-use jobctl.sh runner. Use when: a build/download/model-inference/mass-install takes minutes, a CLI may prompt for input, or the user reports the agent freezing mid-task.

v2.2.4

Quota-aware LLM router that squeezes maximum usable AI out of free-tier API keys across Gemini, Mistral, OpenRouter, Kilo and Cerebras. Probes every model on every key, measures real quality and real published rate limits, then routes each request to the cheapest model that can do the job — spending abundant capacity first and reserving scarce daily quota for when it is actually needed. Persists cooldowns to disk so a 429 discovered in one process is respected by the next. Use when an agent must make many LLM calls on free keys without hitting rate limits, when "all models failed", or when deciding which of several provider keys to use for a task.

v1.0.3

Detects and autonomously parses complex medical DICOM files using structured fallback and verification to overcome standard tool limitations.

v1.0.3

Verify hash-pinned workspace rebuild scripts survive sandbox snapshot wipes, and detect self-mutating config files whose checksum drifts because a later step rewrites an earlier step's output.

v2.0.10

MAX EDITION ethical OS with Seven Lights + Lamp of ehsan + honesty engine calibrated certainty 0-1 + toggle helper heart_toggle.sh + 10-probe self-verification suite detailed probes + per-task playbooks code/research/writing/hard conversations + anti-pattern gallery + stacking integration + anti-tyranny Sunlight Doctrine. Opt-in HEART_OF_LIGHT_MODE=ON.

v1.4.3

Fixes the four things that make a web agent chat feel slow, stale, dumb, or hostile: (1) long laggy "waiting" spinners after you hit send, (2) the agent still answering your PREVIOUS request after a flaky connection recovers, (3) the agent turning into a slow mindless "zombie" late in long conversations, and (4) needless Google/CAPTCHA human verification popping up when you are obviously human, (5) the agent abandoning a TRUE claim just because the user pushed back with displeasure instead of evidence, (6) holding a true claim in the WRONG VOICE - martyred, self-pitying, or joking about something serious - so the correct answer gets discarded anyway, and (7) an agent that either sprays unsolicited cleverness constantly (noise) or invents from nothing instead of from the user's own material. Provides a prompt compactor that reshapes input into a form models prefill 3-4x faster, a request-lifecycle fence that invalidates stale in-flight answers, a context-hygiene monitor that detects and reverses late-conversation degradation, a verification-triggers checklist that reduces bot-score false positives, and an intellectual-spine engine that distinguishes new evidence from mere social pressure so the agent holds true claims under pushback while conceding instantly to real facts (anti-sycophancy, explicitly not contrarianism). Use when a user reports lag, stale replies, degraded quality over time, repeated CAPTCHAs, or an agent that caves and flip-flops whenever it is contradicted.

v1.0.3

Enhances ClawHub skills with structured logging, error recovery, performance monitoring, circuit breaker, and self-healing for robust debugging and stability.

v1.1.6

Tune llama.cpp GGUF inference on CPU-only / edge boxes (1-4 cores, low RAM) for maximum tokens/sec. Complements GPU-oriented tuners. Contains measured, counterintuitive CPU-specific findings — flash attention helps even at short contexts, KV-cache quantization HURTS on CPU, batch size is a no-op, newer-arch Q4 beats older smaller Q5. Use when a model runs without a GPU, on a VPS, container, SBC, or sandbox.

v1.1.3

Red-team auditor for LLM benchmarks with executable scripts, severity calculator, mitigation library, date-contamination detector, evaluator injection harness, and new exploit types (CoT leakage, tool-use gaming). Defensive/research only with proof-of-concept exploits and detection tests.

v1.2.3

Power-user guide to always getting top-tier frontier models on Arena.ai with Max router, Direct vs Agent vs Code selection, rotation caveats, Pineapple mitigation 3-strike, 5-message chunking, local fallback matrix with edge-cpu-gguf-tuner and sandbox-selfheal-guard when Arena down. Updated late July 2026 frontier.

v1.0.3

🎓 Comprehensive career planning for the AI era. Analyze skills, identify opportunities, plan development paths, and navigate career transitions with AI-powered guidance.

v1.6.3

Control Unity Editor via OpenClaw Unity Plugin. Use for Unity game development tasks including scene management, GameObject/Component manipulation, debugging, input simulation, and Play mode control — including arbitrary C# execution (script.execute) and reflection-based editor calls, which can modify scenes, assets, and settings. Use only in trusted local projects; destructive operations (delete, save, package install, code execution) should be confirmed with the user. Triggers on explicit Unity Editor requests like inspecting scenes, creating objects, taking screenshots, testing gameplay, or controlling the Editor.

v1.0.2

Control Unreal Engine Editor via OpenClaw Unreal Plugin. Use for Unreal development tasks including level/actor management, transforms, PIE control, debugging, input simulation, and console commands. Can create/delete actors, save levels, run console commands, simulate input, and capture viewport screenshots and logs — destructive operations should be confirmed with the user. Use only on explicit Unreal Editor requests, in trusted local projects.

v0.6.0

Query Google Search Console, Bing Webmaster Tools, Google Analytics 4 (GA4), and PageSpeed Insights, or run local Lighthouse browser audits, through one self-hosted MCP server. Search analytics queries, summaries, trends, anomalies, comparisons and drop attribution; query and page performance, ranking buckets, cannibalization and opportunity reports; URL inspection, sitemaps, crawl issues, crawl stats, link counts and feeds; GA4 reports covering realtime, content, landing pages, organic search, traffic sources, ecommerce, audience segments, user behavior and conversion funnels; PageSpeed, Lighthouse, and schema checks. Speaks MCP over stdio and Streamable HTTP, plus a FastAPI JSON API. Use when the user wants to inspect SEO, indexing, ranking, browser-audit, or search-traffic data for sites they already own.

v1.0.4

🔐 Secure API key management and rotation for AI agents. Store, encrypt, rotate, and audit API keys used by your agents. Supports multiple providers with centralized key management.

v0.5.5

Operate TraeCN through TRAECNclaw's focused stdio MCP tools. Use when an MCP-capable agent must send or stop work, choose a workspace/model/mode/conversation, inspect or change Trae settings through its UI, manage conversations, or resolve an exceptional question or command approval. Queueing, waiting, recovery, notifications, routine non-command questions, and keep/revert gates are gateway-managed.

v1.0.0

医疗/非医疗内容发布前全链路合规自检：广告法违禁词 + 用户隐私脱敏 + 专家形象核实 + AI生成水印清除 + 平台规则适配 + 受众反应模拟。 统一阀门，一条指令跑完全部检查项（含受众模拟），出问题直接标红。 适用：公众号、小红书、抖音、视频号、培训课件、知识库入库等所有内容发布场景。 触发词：发布前自检、合规检查、违禁词查、内容审核、文案自检、能不能发、广告法检查。

v4.6.0

Nano (XNO) cryptocurrency wallet operations, transaction analysis, and explorer lookups. Use for send/receive, balances, pending funds, address validation, unit conversion, tx/hash/account lookup, explorer links, and Nano block-lattice questions. Prefer xno-mcp first; use xno-skills CLI as fallback.

v5.13.7

Play turn-based arena games over REST

v1.0.7

Through fixed aquarium cameras, the system analyzes fish behavior videos and detects abnormal frictional actions between fish bodies and tank walls, substrate, or rockwork — 'flashing' (fish flipping sideways and brushing tank walls rapidly) and 'scraping' (fish belly/flank rubbing on substrate). The system counts abnormal contact frequency per minute. | 通过鱼缸固定摄像头，分析鱼类的行为视频，检测鱼体与缸壁、底砂、造景石等物体的异常摩擦动作（擦缸：鱼体侧身快速蹭过缸壁；蹭底：鱼体腹部或侧面贴底砂摩擦）。统计每分钟的异常接触频次，当频次超过阈值（默认 5 次/分钟）且持续时间超过 10 秒时，输出'外寄风险提示'，提醒用户检查是否有寄生虫（如小瓜虫、车轮虫、三代虫）感染或皮肤不适。

v1.0.1

网文写作助手（爆款引擎·写审一体）：写小说/创作小说/生成小说/小说大纲/网文创作/番茄小说/起书名/写简介/网文评审/锐评/巨著桌/导出番茄/MD转TXT。含番茄专线、女频、上架文案、内置四十年读者巨著桌、export_fanqie导出脚本。当用户提到写小说、番茄小说、小说大纲、锐评、巨著桌、导出番茄、MD转TXT、番茄粘贴、网文写作助手时启用。