Explore

Apply a simple code transform via agent-booster's WASM engine — sub-millisecond, deterministic, $0 (no LLM call). Companion to cost-booster-route.

Run the corpus benchmark — booster locally, optional Gemini/Sonnet/Opus baselines — and persist a verifiable measured-vs-claimed table

MAD-based outlier detection on session spend. Robust to the very outliers it hunts (unlike mean+sigma). Surfaces specific anomalous sessions with modified-z scores; optional --alert-on-outliers exit code for CI gates. Distinct from cost-burn (aggregate trend) — this answers "which INDIVIDUAL session is the outlier?".

Optimisation findings from your own Claude Code and Codex logs — low cache hit ratio, 1-hour cache writes that did not pay back, sub-agents on a top-tier model, Codex reasoning share, bloated sessions — each with evidence and a what-if saving. Use when asked how to cut spend or why it is high.

Sign, verify, and track fix-marker regressions over time using a deterministic Ed25519 witness manifest. Works in any project — clone the toolkit, run init, register fixes, regen on each release.

Diagnose Ruflo health, then report system, MCP server, and active-agent status without changing the installation

Use when we want to turn a just-finished Formax workflow (e.g. commands, overlays, tools, hooks, permissions, UI parity) into a reusable Codex Skill under .codex/skills, including scaffolding, guardrails, and the minimum test checklist.

Use when preparing a Formax code handoff: selecting files, generating repomix bundles, and writing a high-quality prompt for WebGPT or another coding agent with clear constraints and validation scope.

Safely review, triage, repair, and merge ToolSDK MCP Registry package JSON pull requests. Use when an agent needs to validate files under packages/, detect duplicate registry keys, classify community PRs, make authorized fixes on contributor branches, close invalid or duplicate PRs, or squash-merge approved PRs.

本机文件与代码内容搜索（rg/fd/mdfind/grep 统一入口 seek.py，工具输出即答案）。与 argo 联网搜索互补：搜本机目录、代码符号、文件名、文档全文、文件结构。触发：搜本地、找文件、找代码、本机搜 xx、代码在哪、统计本地命中。

js-run 是 argo 的 JS 运行时车道子技能：不开浏览器跑网页 JS。基于 mini-racer（ISC） + 自研浏览器环境垫片，让「环境探测 + 纯计算」型网页脚本（反爬挑战通行证、混淆数据 解算、JS 内嵌状态提取）在毫秒级、可复现、无真实网络的沙箱里执行。定位是取数链 curl_cffi 与渲染级之间的「算」车道，不替代浏览器（登录态/视觉/真实网络数据不在此车道）。 v0 为独立骨架，未接入主链。Triggers include 跑网页JS、挑战脚本、通行证、JS解密、 逻辑时间、无浏览器执行。

Automatically monitors OKX Flash Earn, Fixed Earn and Flexible Earn opportunities, sends push notifications, and guides subscription. 自动监控 OKX 闪赚、定期和活期赚币机会，推送通知并引导申购。Use when user says: 有闪赚通知我, 监控赚币, monitor earn, notify me about earn, 定时检查理财, 执行 earn-hunter 扫描, earn-hunter scan, 活期年化高了通知我, 监控活期.

Reconstruct editable LaTeX from PDF content using page-aware extraction and visual comparison. Preserve source evidence, flag uncertain math/tables/citations, and distinguish text extraction, OCR, reconstruction, and verified compilation.

Diagnose and repair LaTeX build failures in local projects or supplied logs. Make minimal source fixes, preserve scientific content and reference keys, and report actual build evidence and unresolved errors.

Run when the user invokes /crossreview (or /crossreview:setup to choose the reviewers) or asks for a cross-review, a quorum review or a second opinion from other agents or models: fan a code review (uncommitted work, a branch, a commit range, files, a plan or a document) out to several console code agents (Claude Code, Codex, Coddy, Cursor Agent, Devin, OpenCode, Gemini CLI, Qwen Code, Kimi, Koda) that review it in parallel and blind to each other, then, as the orchestrator, verify every finding against the code and decide alone what matters and what to fix. Works from any agent with a shell on Linux, macOS and Windows; the first run detects the installed CLIs, asks which agents and which of their models to use, and keeps the reviewer roster either for the project (in the agent's folder of the repository) or for all projects (in the agent's home).

A skill with a name that doesn't match the directory.

This is a multiline description. It spans multiple lines. And should be normalized.

A skill that cannot be invoked by the model.

Example skill loaded from resources_discover

Run a lightweight Ultracode workflow for serious coding tasks: plan, split, delegate when useful and allowed by the host, integrate, and verify. Use when the user explicitly invokes ultracode, $ultracode, ultra code, dynamic workflow, workflow orchestration, multi-agent workflow, subagent workflow, parallel agents, swarm, delegate this, split this across agents, or independent verification pass.

An example skill demonstrating the skill file format

Use when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

Use when implementation is complete, all tests pass, and you need to decide how to integrate the work

TDD London School workflow -- mock-first, outside-in test development

Scaffold a Claude Code mod (a plugin with a function-hooks module) from ruflo's governed template — hybrid hooks.json with a classic fallback, a host adapter over literal $ calls, userConfig options, engine-kit tests and a tsconfig — then validate and test it. Use when the user asks for a mod, a function hook, or wants Claude Code to modify itself.

Fetch metadata and audio_url for a single production by id

Orchestrate multi-phase deep research with web search, memory retrieval, pattern matching, and synthesis into structured findings

Use when executing implementation plans with independent tasks in the current session

Run health checks on the Ruflo installation and fix common issues

Initialize a new Ruflo project with MCP tools, hooks, and agent configuration. Use when setting up Ruflo in a fresh repo, or when the user says "init ruflo", "set up ruflo", or asks how to bootstrap the MCP server, hooks, and agent configs from scratch.

Discover and recommend ruflo plugins based on your workflow, installed MCP tools, and current task

Generate a data-only animated theme preset for Ruflo Console Theme Studio. Use when the user requests a new console color palette, cyberpunk theme, or variation of Orbital globe, stars, and swarm animation in Claude or Codex.

Run one tick of the sales business-pod (ADR-164 §4.1, Phase 2). Loads templates/sales.json, validates it against the pod-schema, resolves agents against ruflo's agent registry, reserves budget via the Phase-2 file-based stub ledger (atomic SQLite tracker is Phase 3 per ADR-164.1), constructs per-agent dry-run prompts, posts a summary envelope to room "sales" via the federation_bbs_publish JSONL backing store, and emits a structured {podName, tickId, agentsRan, totalUsd, envelopeId, status} line for /loop ingestion. Dry-run by default; --live is reserved for Phase 3.

UI test recipe -- composes browser-record (capture) + browser-replay (verify) so every test produces a replayable RVF artifact, not an ephemeral run

Visual + DOM diff between two recorded sessions at matching trajectory step ids; used for visual regression and replay verification

DEPRECATED in v0.2.0 -- use browser-extract instead; this is a thin shim for backward compatibility, kept as a shim and slated for removal in the next minor release

Replay a recorded session trajectory against the same URL or a mutated variant; uses browser-selectors embedding similarity to recover from DOM drift

Open a named, traced browser session into an RVF cognitive container with a ruvector trajectory recording every action

Drive an authentication flow once, sanitize cookies through AIDefence, and vault a reusable cookie handle in browser-cookies for future sessions

Execute a natural-language browser intent via page-agent (browser_act) when the target is easier to describe than to select — degrades gracefully when page-agent or an OpenAI-compatible LLM provider isn't configured

Fill a web form by mapping field-name → value, with optional template lookup from browser-templates for known forms

Extract structured data via stored browser-templates or one-shot DOM queries, with mandatory AIDefence PII + prompt-injection gates before content reaches the model

Probe a site's authentication flow for redirect leaks, missing CSRF, weak session cookies, and OAuth misconfiguration; produces an auth findings.md

Join and operate a signed cross-host agentbbs federation, and coordinate work claims across nodes. Use when: connecting ruflo agents across machines, sharing status/tasks/results between hosts, propagating work claims across a swarm, or standing up a federation hub. Skip when: single-host local work with no other nodes to coordinate with.

Use learned patterns and current state to predict the optimal next action

Run an autonomous /loop iteration -- check progress, work on next task, schedule next wake

Scan inputs for prompt injection, unsafe content, and adversarial attacks using AIDefence. Use when processing untrusted input (user submissions, API payloads, webhook data, tool outputs) before passing it to a model or executing it.

Detect and flag personally identifiable information (PII) in text, code, and configurations. Use before committing code, writing logs, storing data, or sending model responses that might contain emails, phone numbers, SSNs, API keys, or passwords.

Checks an AI team run against its acceptance criteria using the evidence bundle. Use when the user asks whether the work is done or correct, or before completing a run. Separates verified evidence from agent claims.

Stores approved context for one AI team and searches it later, with provenance. Use when the user asks to remember a decision, constraint, or fact for the team, or to recall what the team already knows. Memory is separate for each team and tenant.

Reports an AI team's progress, blockers, task ownership, and recorded budget. Use when the user asks for status, what is blocked, who owns what, or how much of the budget is recorded as used. Read-only.

Pauses, resumes, or completes an AI team, or marks a run complete, while keeping its evidence. Use when the user asks to pause, stop, resume, wrap up, or close a team or run. Changes coordination state only and deletes nothing.

Plans and creates an AI team for a goal, with roles, a run that carries a unit budget, and bounded tasks. Use when the user asks to start, set up or staff a team, or to split a larger goal across roles. Records coordination state only; nothing is executed.

Creates, assigns, and updates tasks in an AI team run so work has one owner and one deliverable each. Use when the user asks to break down, assign, claim, block, or complete tasks in a run. Changes recorded task state only.

Show AGNTCY/SLIM/CASA integration status — whether the optional SLIM runtime package is installed, whether a SLIM endpoint is configured, which transport is the default, and what the agntcy mod has seen this session. Use when the user asks "is AGNTCY configured?", "show SLIM/CASA status", or "is AGNTCY/IOC integration active?". Read-only; makes no network call.

Vector search via embeddings_* (large-scale HNSW) and ruvllm_hnsw_* (WASM router for ≤11 hot patterns), with RaBitQ 1-bit quantization for 32× memory reduction

Query AgentDB through the controller bridge -- semantic routing, hierarchical recall, causal graphs, context synthesis, pattern store/search

Browse, publish, and install WASM agents from the community gallery

Create and manage sandboxed WASM agents for isolated code execution