AI & ML

getedgehq Bundle

Mine agent session logs (Claude Code, Codex, OpenCode) for user corrections, recurring failures, user frustration, token-heavy sessions, and repeated manual workflows, then find and score skills that could fix them. Use when asked "what skills am I missing", "mine my session logs", "where do my agents keep failing", "find a skill for X", "which sessions burned the most tokens", or to discover existing skills from skills.sh / Floom / local pools before writing anything new. Pairs with skill-eval-loop, which proves candidates before adoption.

getedgehq Bundle

Run a credible with-skill versus without-skill evaluation for an Agent Skill, preserve evidence, and prepare a public-safe result. Use when the user asks whether a skill actually works, wants to benchmark a skill, compare a baseline with a skill, or generate proof for a skill claim.

getedgehq Bundle

Create a compact, evidence-backed proof-of-work receipt for a completed AI agent run. Use when the user wants to summarize what an agent actually built, how long it took, what changed, tests run, interventions, deployment status, or skills used.

getedgehq Bundle

Diagnose and fix an unreliable, expensive, or unsafe LLM agent by auditing its harness (golden set, judge, cost caps, data layer, action approvals, tracing) before blaming or swapping the model. Use when someone says an agent is "burning tokens", "hallucinating", "brittle", gives inconsistent answers, asks whether to switch to a cheaper/better model, or wants to ship an agent or prompt change to customers.

getedgehq Bundle

Convert messy project notes, meeting transcripts, or agent logs into an evidence-linked ledger of decisions, reversals, open questions, owners, and next actions. Use when the user asks what was actually decided, wants a decision log, or needs a reliable handoff from an unstructured discussion.

getedgehq Bundle

Prove whether a candidate skill actually improves agent performance before adopting it - blind A/B evals on reconstructed real tasks, a three-tier adoption gate, and production rechecks. Use when asked to "eval this skill", "does this skill help", "test this skill before installing", "run the auto-improve loop", or "recheck probationary skills". Takes candidates from skill-miner; runs locally with the user's own agent CLI and models.

getedgehq Bundle

Analyze Claude or Codex session logs or pasted prompts to identify evidenced strengths, weak task categories, a surprising weakness, and up to seven relevant skills, then create a private JSON result and shareable card. Use when asked for an AI skills gap test, agent résumé, missing expertise, or a share card about what an agent is good or bad at.

getedgehq Bundle

Test and improve whether an agent invokes a skill on the right requests and avoids invoking it on near-miss requests. Use when a skill works explicitly but is under-triggered, over-triggered, or has a weak description.

getedgehq Bundle

Audit, optimize, and benchmark a user-owned system prompt, developer prompt, CLAUDE.md, AGENTS.md, or agent instruction file against representative tasks. Use when the user wants to improve an agent's core instructions, make an agent more reliable, reduce instruction bloat, fix tool-use or ask-vs-act behavior, or prove whether a revised system prompt is actually better.

getedgehq Skill

Build repeatable evaluations for an agent from production failures, deterministic verifiers, cost, latency, and trace evidence.

getedgehq Bundle

Check whether one additional AI-agent Skill would materially help with a task, then find the best matching GetEdge Skill and return its exact install command. Use when someone asks which Skill to use, says "find a skill for this", invokes SkillNeed before starting work, or wants to know whether their current agent setup is missing relevant procedural expertise.

zerotokenclaw Bundle

Unitree robot software stack: which SDK, repo, or package applies to a given Unitree robot, CycloneDDS and network setup, install paths, and legacy-SDK traps. Use whenever writing or reviewing code that talks to Unitree robots (G1, G1D, Go2, B2, H1, H2), or when choosing ROS/ROS2 packages, simulators, or model assets for them.

zerotokenclaw Skill

Use when entering read-only plan/design phase before coding: explore codebase, draft plan, get user approval, then exit to implement. Trigger on 计划模式、 EnterPlanMode、ExitPlanMode、先规划再写代码、plan mode。

zerotokenclaw Bundle

Use when entering an unfamiliar project to analyze it and generate a project-specific skill in the project's .{agent}/skills/ directory

zerotokenclaw Skill

Use when creating custom coding subagent specs as JSON: identifier, whenToUse with examples, systemPrompt. Trigger on 创建agent、自定义子代理、agent JSON、whenToUse、agent architect。

zerotokenclaw Bundle

Route to Chinese vertical skills sourced from anbeime/skill store: content publishing, short video, ecommerce, TTS/ASR, digital avatar, arch diagrams, multi-agent teams, legal docs, stock analysis, Obsidian, resume. Trigger on 内容创作、公众号、小红书、短视频、电商、带货、TTS、配音、数字人、架构图、 archify、多智能体、合同审核、论文分析、Obsidian、简历、技能商店、anbeime。

zerotokenclaw Bundle

Use when designing or implementing embodied-AI / robot voice pipelines (Wake/KWS → VAD → ASR → LLM/Agent → TTS → Actuator), choosing ASR/TTS/VAD stacks for ROS2/Jetson/RISC-V/edge robots, or integrating barge-in, AEC, and tool-calling with motion. Trigger on 具身智能语音、机器人语音、ASR+LLM+TTS、 ROS2 voice, EdgeVox, FunASR, CosyVoice, sherpa-onnx, Piper.

nevamind-ai Bundle

Install or uninstall memU for whatever agent you are — identify your host, print its packaged guide, and follow it to wire (or unwire) both seams (record and inject). Use when the user asks to install, set up, integrate, remove, or uninstall memU.

frederico-kluser Bundle

Multi-agent research orchestrator on Brave Search and nothing else. The main agent never searches: it raises every doubt, fires a burst of parallel sub-agents (one closed question each, at most 10 at a time; sub-agents=N), then checks whether the answers opened new questions — single-burst, or continuous-burst until saturation. A context burst reads the conversation and the repo before any web search; each sub-agent uses the surf-ai CLI and returns a structured handoff. Stops at exit 78 (no valid Brave key): no fallback, no free tier. Use when answering needs MORE THAN ONE dependent question. Triggers: ache tudo sobre, levantamento completo, pesquisa profunda, panorama de, compare X, Y e Z, deep dive, find everything about, exhaustive research. NOT for ONE verifiable answer (version, number, date, price, two options on one axis) — that is surf-search-agent-skill; not for plans (surf-plan-agent-skill), local files, git, code, or reading a URL (Brave returns links and snippets, never page content).

frederico-kluser Skill

Generates a research-grounded execution plan for a coding task. MUST BE USED whenever the user asks for a plan, design, architecture or spec — including in plan/approval mode, BEFORE any plan is presented for approval. Reads the project, runs MANDATORY web research (surf-search-normal / surf-ai via Bash — Brave Search only, no fallback; WebSearch/WebFetch only when the harness itself denies Bash, never for a missing or burned key), interviews the user with research-backed options, then delivers a plan with cited sources and a research ledger. Vague, high-stakes or hard-to-reverse work — or "raise all my doubts first", "exhaustive plan", "levante todas as dúvidas", "plano exaustivo" — switches to Deep mode: a full ambiguity sweep before any question. Triggers: "make a plan", "plan this", "design…", "architect…", "spec this out", "faça um plano", "planeje isso", "monte um plano", "arquitete". NOT for trivial edits — only work that warrants a written plan (≥30 min, ≥3 files, or an architectural decision).

frederico-kluser Skill

Project knowledge map for surf-agent-skill — an autonomous web research tool for AI coding agents. Maps the repo structure, skills, conventions, and key files.

frederico-kluser Skill

Web search for ONE question, answered and cited, with no orchestration: one surf-search-normal call (Brave Search, nothing else) returns the answer with [n] citations and a source table — no sub-agents, no doubt ledger, no files, no commit. Use for ONE independent question with ONE verifiable answer: a number, version, date, limit, price, error message, "is this still true?", "how do I do X", two options on a SINGLE axis, or a plain list of links. Triggers: pesquise, procura na web, me acha, qual a versao de, ainda existe, quanto custa, o que e, como faco, search the web, look this up, what is, how do I, does X support. Brave only: without a VALID key every command exits 78 and this skill STOPS — no fallback, no WebSearch reserve. NOT for MORE THAN ONE dependent question, a levantamento/panorama/deep dive, 3+ options or several axes — that is surf-research-agent-skill. NOT for plans (surf-plan-agent-skill), local files, git, code, or reading a URL (Brave returns links and snippets, never page content).

zerotokenclaw Bundle

SpacemiT Robot SDK entry point for vision and image development. Use when users ask to set up, build, run examples, call APIs, benchmark, or debug detection, classification, segmentation, tracking, pose, face recognition, OCR, image-text embedding (CLIP/SigLIP), or composed CV applications (fall/fire/intrusion/emotion detection) in local, remote, or hybrid modes.

zerotokenclaw Bundle

Use when building LangGraph / multi-agent coding systems inspired by Claude Code-style tool loops: plan mode, Explore/Plan subagents, Task fan-out, agent JSON specs, CLAUDE.md context. Trigger on LangGraph、编码Agent、计划模式、 子代理、Explore、Plan、EnterPlanMode、创建agent。

zerotokenclaw Skill

Implements secure and accurate facial recognition workflows including embedding comparison, validation, and check-in logic.

zerotokenclaw Skill

Use when launching Explore/Plan/general-purpose subagents via Task-style fan-out: parallel research, readonly explore, detailed prompts, summarize results to user. Trigger on 子代理、Explore、Plan、Task工具、并行agent、 subagent。

altslate-labs Bundle

Find "jevable" code in a repository: existing semantic decisions that could be expressed as Jev Choice, Score, or Noul questions. Use when asked to audit a codebase for Jev opportunities, replace brittle semantic heuristics or LLM classification/judging calls, identify typed decision boundaries, or plan a Jev migration. Produce ranked, source-grounded findings with concrete question contracts, batching/dependency analysis, fallback behavior, and a validation plan. Default to reviewing code; implement only when requested.

gabrielmoreira Skill

Run a multi-agent review of changed files for scope, reuse, quality, efficiency, clarity, and altitude issues followed by fixes. Use when the user asks to "simplify code", "review changed code", "check for code reuse", "review code quality", "review efficiency", "simplify changes", "clean up code", "refactor changes", or "run simplify".

gabrielmoreira Skill

Create a pull request using repository branch rules, title conventions, templates, and concise English descriptions.

gabrielmoreira Bundle

Diagnose a reproducible failure, fix its cause, and verify the regression. Use for crashes, incorrect behavior, and failing tests.

gabrielmoreira Bundle

Summarize AI conversation histories for a specified date or period. Use for daily work recaps and cross-tool activity summaries.

gabrielmoreira Bundle

Find external documentation, web sources, or remote code with citations. Local code navigation uses the configured code tools.

gabrielmoreira Skill

Use when the user asks whether a skill exists, wants help finding or installing one, or asks how to extend agent capabilities with a reusable skill. Start with `npx skills find`; if that fails, use a GitHub-aware fallback before concluding no good match exists.

gabrielmoreira Skill

MCP Server 开发指南 - 帮助创建高质量的 MCP (Model Context Protocol) 服务器。 当需要构建 MCP 服务器来集成外部 API 或服务时使用，支持 Python (FastMCP) 和 Node/TypeScript (MCP SDK)。 触发关键词: mcp开发, mcp服务器, model context protocol, 构建mcp, 创建mcp

gabrielmoreira Skill

Run code review in an independent agent. Automatically used after code implementation or fixes. Triggered by "review this", "code review", "code check", "look at this PR", "check the diff", "is this code OK?".

gabrielmoreira Bundle

Use when PDF files are the primary input or output: read, compare, reconcile, extract text/tables/images, OCR scans, fill forms, split, merge, rotate, rename, compress, or convert between PDF and images. Optimized for private financial, tax, legal, and health documents on macOS.