运行真实任务基准测试套件，对 OpenClaw 智能体进行测试，从而在模型或框架变更投入生产工作流之前进行对比。
该技能提供了一个名为 PinchBench 的基准测试套件，用于在部署前评估 OpenClaw 编程智能体的性能。通过运行可重复的真实任务，开发者可以量化比较不同模型或框架变更对智能体执行实际编码任务能力的影响。该技能支持 OpenClaw 实例，依赖 Python 3.10 及以上版本、uv 包管理器以及 PinchBench 仓库。模型 ID 必须包含提供者前缀（如 openrouter/ 或 anthropic/），默认路由提供者为 OpenRouter。此工具对于在关键生产工作流中部署前，验证智能体调用正确工具和参数的能力至关重要。
npx skills add https://github.com/agentskillexchange/skills/tree/main/skills/benchmark-openclaw-coding-agents-against-repeatable-real-tasks-before-rollout-with-pinchbench运行真实任务基准测试套件，对 OpenClaw 智能体进行测试，从而在模型或框架变更投入生产工作流之前进行对比。
正在运行的 OpenClaw 实例，Python 3.10+，uv，PinchBench 仓库副本，以及上游文档中说明的模型提供者凭据
使用与您环境匹配的上游安装或设置路径：
上游的要求和注意事项：
基本用法或入门说明：
工具使用 — 模型能否使用正确的参数调用正确的工具？
bash
从上游文档提取：https://raw.githubusercontent.com/pinchbench/skill/HEAD/README.md
This repo contains 20 individual skills — each has its own dedicated page.
Builds citation networks from Semantic Scholar API and CrossRef DOI metadata. Visualizes paper influence graphs using NetworkX, identifies seminal works, and tracks research lineage across fields.
Exposes Advanced Custom Fields data through the WordPress REST API using register_rest_field and acf_format_value. Handles repeater fields, flexible content layouts, and gallery fields with proper serialization.
An ASE skill built around ACF Extended, the WordPress enhancement suite for Advanced Custom Fields that adds field types, admin improvements, front-end forms, options pages, and developer tooling. It is a practical fit for agents working inside complex WordPress content models and custom field workflows.
Converts Advanced Custom Fields field groups into native Gutenberg blocks using the ACF Block API v2 and @wordpress/scripts build pipeline. Maps ACF repeaters, groups, and flexible content to InnerBlocks and block attributes with server-side rendering via acf_register_block_type().
Activepieces is an open-source, self-hostable workflow automation platform with 200+ integrations. It provides a visual builder for creating automated workflows and exposes all its connectors as MCP servers for AI agent use.
ActivityWatch is a privacy-first, open-source automated time tracker that records application usage, browser activity, and AFK status across Windows, macOS, and Linux. With 16k+ GitHub stars, it provides detailed productivity analytics without sending data to external servers.
act is an open-source CLI tool that runs GitHub Actions workflows locally using Docker, enabling fast feedback on workflow changes without pushing to GitHub. It is a standard tool for local Actions development and testing.
Use AgentClick when an agent should pause before risky commands, plans, drafts, or code changes so a human can inspect, edit, approve, or reject them in a purpose-built browser UI.
Use UX/UI Agent Skills when Claude should generate tokens, component specs, accessibility audits, and framework-specific UI code from a repeatable design workflow.
Use Bats-core when an agent needs to turn fragile shell scripts or command-line workflows into something it can verify repeatedly after edits. The agent writes focused Bash tests for success paths, failure paths, and output contracts, then runs them locally or in CI before a refactor, release, or incident fix goes out.
Use Cognee to ingest project knowledge into graph and vector memory so agents can retrieve durable context across sessions and workflows.
Give a coding agent symbol-aware lookup, cross-file rename, and structural edit tools before it starts making brittle text-only changes.
Give LangGraph agents memory management and search tools so they can store, retrieve, and update durable facts across sessions.
Use SimpleMem to store, compress, index, and retrieve text or multimodal memories for agents through MCP or Python integrations.
Keep Claude Code sessions grounded in prior decisions, project context, and daily handoff notes instead of starting from zero every time.
Store embeddings beside application data in Postgres, create vector indexes, and query nearest neighbors for semantic search, RAG, recommendations, or agent memory retrieval.
Find the open PR for the current branch, gather unresolved review comments, and drive a focused comment-resolution workflow with gh-authenticated context.
Use VoltAgent to intercept, validate, and enforce input/output policies in TypeScript agent workflows.
Use Zep as an external context layer for agents that need to store events, assemble temporal graph context, and retrieve relevant memory before model calls.
Calls Adobe Firefly
Extract frames or short clips from videos using ffmpeg.
Summarize or transcribe URLs, YouTube/videos, podcasts, articles, transcripts, PDFs, and local files.
Use when testing, fixing, or extending the OpenClaw Control UI GUI with Vitest + Playwright end-to-end checks, mocked Gateway WebSocket flows, mocked dashboard runs, screenshots/videos, or agent-verifiable browser proof.
Obra superpowers agent skills.
See, Understand, Act on video and audio. See- ingest from local files, URLs, RTSP/live feeds, or live record desktop; return realtime context and playable stream links. Understand- extract frames, build visual/semantic/temporal indexes, and search moments with timestamps and auto-clips. Act- transcode and normalize (codec, fps, resolution, aspect ratio), perform timeline edits (subtitles, text/image overlays, branding, audio overlays, dubbing, translation), generate media assets (image, audio...
End-to-end marketing campaign planning and execution. Covers audience research, positioning, campaign angle definition, landing page copy, email sequences, social posts, ad copy, short-form video scripts, and content calendars. Use as the orchestration layer for multi-channel product launches. Use when planning or executing a multi-channel product launch, or producing landing page, email, social, or ad copy.