globalcaos 发布的技能

为 OpenClaw 智能体加上 JARVIS 风格的英式嗓音与冷面幽默，一条命令同步输出语音和紫色聊天气泡。

OpenClaw 原生 WhatsApp 通道：22 项消息与群组操作，叠加 Protocol v2 多智能体协同。

一个本地面板，统一追踪 Anthropic、Gemini、OpenAI、Manus 四家提供商的 token 用量与预算告警。

Stop sending 'format this JSON' to Opus. Stop sending 'cron job' to GPT. Billing-aware routing guide for choosing among the models already configured in your OpenClaw setup when assigning an agent, sub-agent or cron task — flat-rate first, metered only when justified, budget pressure respected. Not for picking models outside your configuration, and not a runtime proxy.

Your agent says 'done' — but did it check? Superpowers turns any OpenClaw agent into a disciplined engineer. Verification iron law (evidence before claims), three-agent code review (build → verify spec → verify quality), systematic debugging (4-phase root cause, three-strike rule), brainstorming gates (design before code), and anti-over-engineering rules. Use when: (1) coding tasks of any complexity, (2) debugging failures, (3) about to claim work is complete, (4) spawning sub-agents, (5) planning features, (6) reviewing code. Inspired by top coding agent methodologies, adapted for OpenClaw multi-agent architecture.

Read and search Outlook, inspect attachments, and create or edit drafts without any send endpoint. Uses one short-lived Microsoft Graph access token supplied on stdin for one run; it never stores credentials. Bulk mailbox export is opt-in.

12 humor patterns for AI agents based on embedding space bisociation theory. Operational reference for generating contextually appropriate humor during conve...

You spawned 4 sub-agents. One died 20 minutes ago. You're still waiting. Overseer watches them so you don't have to — zero tokens, pure OS-level process checks. No polling loops, no wasted heartbeats.

Be one of the first to benchmark your agent's memory — and help shape how AI remembers. Peer-review-grade evaluation (LLM-as-judge, nDCG/MAP/MRR with 95% CIs, ablations) against your live memory system. Runs entirely LOCALLY by default — no memory content leaves your machine, and excerpts are redacted even on the local path. The optional OpenAI judge is opt-in, prints the exact request body it would send, redacts secrets first, requires typed consent, and cannot be switched on by an unattended run. Submitting results is a separate confirmed step that validates the report against the full schema and previews every field in it, and identifies you only if you pass --contributor. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. See Permissions, Data Flow & Consent.

Free transcripts, 4K downloads, and video exploration — zero API quotas burned. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. Search and your own account data run on YOUR Google OAuth client, read-only and opt-in, with a one-command logout — see Permissions, Data Flow & Consent.

Three env vars. One script. Your agent manages your WordPress site — and cannot quietly change it. Reads are free and new posts land as drafts; anything visitors would see (publishing, editing or trashing live content, uploading media, moderating comments) needs WP_ALLOW_PUBLISH=1 on that call, and plugins, themes, users, settings and any route the script does not know need WP_ALLOW_ADMIN=1. WP_URL is parsed strictly and credentials go only to that HTTPS host (pin it with WP_ALLOWED_HOSTS). Plugin install is code execution on your site and is named as such. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. See Permissions, Data Flow & Consent.

Classify every shell command as SAFE, WARN, or CRIT before your agent runs it. The classification is instruction-only and runs nothing. The package also ships optional installer scripts that MODIFY SOURCE CODE in an OpenClaw checkout you point them at — they refuse non-OpenClaw trees, verify the whole edit in a temp file before touching the real one, back up, require --yes, offer --dry-run, never rebuild unless you ask, and ship with an unpatch off-switch that validates its range instead of deleting between markers. Nothing is patched by installing this skill. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. See Permissions, Data Flow & Consent.

Stop guessing what your AI costs. Tinker shows every token, every dollar, every context byte — in real time.

Claude follows your rules. GPT ignores half of them. Gemini invents new ones. Model Prompt Adapter patches the gaps — per-model addenda that fix scope creep,...

Your agent crawls LinkedIn through the browser session you already have — profiles, search, connections, inbox, feed. No official API, no app review. Your cookies never leave the browser: API calls are a fetch() run inside the linkedin.com tab you shared (people-search navigates that tab to a LinkedIn search page and reads the results), and 1.2.1 DELETED the cookie-extraction, session-store and external-replay code from the package rather than leaving it switched off — there is no longer anything to store or steal. Every request, navigation and tab pick is pinned to exactly https://www.linkedin.com, including the LINKEDIN_TARGET_ID override, so a look-alike host cannot borrow your session. Reads and drafts freely; the one write, message-send, needs per-action consent that repeats the exact conversation URN, and without it you get the draft and nothing is sent. The browser relay must be a literal loopback IP, with no override. Daily ceilings count every record requested and are reserved

Your agent crawls LinkedIn through the browser session you already have — profiles, search, connections, inbox, feed. No official API, no app review. Your cookies never leave the browser: API calls are a fetch() run inside the linkedin.com tab you shared (people-search navigates that tab to a LinkedIn search page and reads the results), and 1.2.1 DELETED the cookie-extraction, session-store and external-replay code from the package rather than leaving it switched off — there is no longer anything to store or steal. Every request, navigation and tab pick is pinned to exactly https://www.linkedin.com, including the LINKEDIN_TARGET_ID override, so a look-alike host cannot borrow your session. Reads and drafts freely; the one write, message-send, needs per-action consent that repeats the exact conversation URN, and without it you get the draft and nothing is sent. The browser relay must be a literal loopback IP, with no override. Daily ceilings count every record requested and are reserved

Discover all inbound links (backlinks) to a domain, subdomain, or GitHub repo, then classify each as "ours" (we created/control the source) vs "organic" (someone else). Use when the user asks to find/audit backlinks or inbound links to a site, check who links to a domain or a GitHub repo, separate self-made links from organic ones, or refresh an inbound-links graph. Wraps four sources (GitHub repo referrers, a list of URLs you found, the backlinks.sh Common-Crawl API, and a Google Search Console CSV export) behind one classify-and-report CLI. Two of the four need no account at all; the one optional API key is stored in your OS keychain and cleared by --logout. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. See Permissions, Data Flow & Consent.

Export your ChatGPT conversations to local files — titles, timestamps and, if you ask for it, full message text. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. Invoke it BY NAME — "chatgpt-exporter-ultimate: export my conversations" — so ordinary talk about ChatGPT cannot trigger it. Rides the ChatGPT session your browser already holds — browser relay or a bookmarklet you paste yourself. It reads NO access token, no cookie jar and no credential file. The relay enumerates your conversation list and writes JSON and Markdown copies to a private directory; it defaults to index-only (titles and timestamps, no message text) and refuses to run without an explicit confirmation of destination and scope. The bookmarklet can also find conversations inside Projects, by running searches against your history, and downloads one JSON file after three dialogs. Relay destinations are resolved through symlinks before anything is written, so a relay export cannot land in a synced folder

Render comparison tables a human reads at a glance instead of parsing — colour and bar length carry the comparison, text only confirms it. Use when you are about to present several options with several attributes each (search results, product comparisons, model benchmarks, server health, candidates, quotes) in a chat that renders HTML, or when asked to turn such a comparison into a clearer table. Not for a single value, a plain list, or a channel that does not render HTML. Ships a five-tier colour vocabulary, chip, bar and number primitives, a renderer that escapes every value it is given, and the per-domain column-spec pattern.

Discover all inbound links (backlinks) to a domain, subdomain, or GitHub repo, then classify each as "ours" (we created/control the source) vs "organic" (someone else). Use when the user asks to find/audit backlinks or inbound links to a site, check who links to a domain or a GitHub repo, separate self-made links from organic ones, or refresh an inbound-links graph. Wraps four sources (GitHub repo referrers, a list of URLs you found, the backlinks.sh Common-Crawl API, and a Google Search Console CSV export) behind one classify-and-report CLI. Two of the four need no account at all; the one optional API key is stored in your OS keychain and cleared by --logout. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. See Permissions, Data Flow & Consent.

Your phone's location, on your own machine, behind a token. An OwnTracks HTTP receiver that stores the trail in a local SQLite/JSON store you own, with named places and distance queries. Binds to 127.0.0.1 by default and refuses to start without a token, so there is no unauthenticated path to your location — reads and writes both need it. Retention is capped by default (7 days / 500 points) and `--purge-data` erases the trail on demand. Use when: (1) "where am I?", (2) storing named locations (home, gym, work), (3) querying nearby places by distance, (4) checking location history. Requires: OwnTracks app, Node.js 22+, better-sqlite3. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. See Permissions, Data Flow & Consent.

Amazon shopping that ends in a decision, not a page of links. It sweeps a dozen query phrasings instead of trusting one keyword, decodes the spec markings a listing hides (memory-card A2/V30 classes, active-ingredient concentration), and ranks on the metric that actually decides the buy — €/kg of active, €/GB, €/kg of protein — so the cheapest sticker price stops winning by default. Availability is a hard gate: a listing that still renders is not a listing you can buy. It reads amazon.es ANONYMOUSLY — there is no login, no cookie capture, no browser tab to share and no stored session, because 1.2.1 removed that code from the package rather than leaving it switched off. It reads no credential of your own; the only secrets it touches are API keys you set yourself for two optional paid paths (Apify, the Amazon Creators API), and with neither set it makes no third-party call. It searches amazon.es and nothing else — the classifieds scrapers earlier versions shipped behind a flag are gone.

Reference reading, not a runtime policy: a field guide of 40 lessons from running AI agents 24/7 on real tasks — ethics, messaging security, context and memory layout, configuration safety, scheduled jobs, multi-model routing, budget, bot collaboration, epistemic hygiene, and how to improve checklists over time through human-reviewed proposals. Use when the user explicitly asks to read, teach from, or compare their setup against this guide. It installs nothing, schedules nothing and writes nothing. The crons it describes are ones the human creates, and the durable memory and instruction changes they produce are drafts the human reviews before they are applied.

Reference material, not an executable tool: one document (GUIDE.md) with ten concrete steps for cutting OpenClaw token consumption on a Claude subscription — heartbeat model, session compaction, cron model and frequency, bash-ified crons, workspace file size, cache retention. Read it when asked why token usage is high or how to lower it.

Reference material, not a runnable skill: a one-page outline of a workflow for scanning GitHub forks and ClawHub skills (bash pre-filter, sub-agent fan-out, scheduled reports). It ships no scripts and has no entry point. Read it only for the workflow shape; for working scanner code use the sibling skills/fork-scanner/ directory.

Stop editing files one at a time. ORCA drafts every change in parallel, then applies them per-file-serialized so disjoint files fly and shared files never collide. It SPAWNS SUBAGENTS on your own provider (one per unit, which costs money) and writes ONLY the repo-relative files you list. Patches that name any other file are rejected in code. Because a symlinked path component can carry a write outside the repository, a separate checker agent resolves every path before any writer is spawned and again after apply, and lists the change set git can see; code refuses or fails the units it flags. The writers and checkers are ordinary subagents running as your user: the checkers are told not to write and their output is self-reported, and git cannot see writes outside the repository — this is detection, not a sandbox. Committing rewrites git history and needs commit AND confirmedCommit; it is OFF by default and never inferred. In its default mode it creates a git worktree and branch per unit-

Export your ChatGPT conversations to local files — titles, timestamps and, if you ask for it, full message text. Built for the TinkerClaw fork — github.com/globalcaos/tinkerclaw. Invoke it BY NAME — "chatgpt-exporter-ultimate: export my conversations" — so ordinary talk about ChatGPT cannot trigger it. Rides the ChatGPT session your browser already holds — browser relay or a bookmarklet you paste yourself. It reads NO access token, no cookie jar and no credential file. The relay enumerates your conversation list and writes JSON and Markdown copies to a private directory; it defaults to index-only (titles and timestamps, no message text) and refuses to run without an explicit confirmation of destination and scope. The bookmarklet can also find conversations inside Projects, by running searches against your history, and downloads one JSON file after three dialogs. Relay destinations are resolved through symlinks before anything is written, so a relay export cannot land in a synced folder