AI Agent Workflow Collection

Extracts pure article text from a raw text file, HTML dump, or URL. Use this skill whenever the user asks to clean up content, remove ads/navigation from text, or extract the main article from a noisy source.

Shared content summarisation and AI analysis pipeline. Provides reusable references for thesis-driven analysis quality rules, AI analysis with user-preference calibration, and suggestion logging. Referenced by ingest skills (ingest-newsletter, ingest-threads, ingest-youtube, ingest-website). Do not invoke this skill directly.

Reviews today's reports in ./reports/ and generates a synthesized Knowledge Distillation document in reports/distillations/. Use this skill whenever the user asks to "summarize today's reports", "distill today's knowledge", "run daily distillation", or "update the knowledge distillation document". It ensures technical insights about AI Agents and Context Engineering are preserved and systematized.

Execute CLI commands and technical workflows inside a project's dev-container, ensuring environment consistency with project-defined dependencies and tool versions. Use this skill whenever the workspace has a .devcontainer/ directory and the task requires running shell commands, build tools, test suites, or any CLI tools. Trigger even if the user doesn't say "container" — if they ask to run tests, execute a workflow, or use a project-specific CLI and a .devcontainer exists, bridge into the container first. This prevents "works on my machine" issues and ensures the right runtime versions are always used.

Generate or update an Evolution Log — a narrative document that tells the development history of a project through iterative problem-solving cycles (Problem → Options → Decision → Result → New Problem). Use this skill whenever the user says 'update my evolution log', 'generate evolution log', 'write the project history', 'document the development journey', 'update EvolutionLog', 'add recent changes to evolution log', 'chronicle the project evolution', or any request to create or maintain a narrative development history from git commits, RFCs, RCAs, ADRs, or other decision documents.

Fetch all content from a Threads (threads.net / threads.com) post URL using agent-browser. Use when the user provides a Threads post URL and wants to extract the full post text, author info, media URLs, engagement metrics, and reply thread. Triggers: 'fetch this Threads post', 'scrape Threads', 'get content from Threads URL', 'extract Threads post', or any request involving a threads.net/threads.com post link.

Full PR lifecycle via isolated worktree, atomic commits, PR creation, and a strict verification loop (CI + review-work) until merged.

Fetch a Threads post and produce a summary report in the configured output language (defaulting to English), then mark the Google Task as completed. Use when the user provides a Threads URL to process, says 'process this Threads post', or when invoked by daily-workflow for tasks in the Delegate list.

Fetch a generic website URL and produce a summary report in the configured output language (defaulting to English) using the Jina Reader API, then mark the Google Task as completed. Use when the user provides a website URL to process, says 'ingest this article', 'summarize this page', 'process this website', or when invoked by daily-workflow for tasks in the Delegate list that are not Threads or YouTube URLs. Triggers include any http/https URL that is not threads.net, threads.com, youtube.com, or youtu.be.

Automatically process URL-based tasks from the Google Tasks 'Delegate' list. Supports Threads posts (threads.net / threads.com) and YouTube videos (youtube.com / youtu.be). Fetches content, generates a Traditional Chinese summary with verbatim raw content appended, saves a Markdown report, and marks the task as completed. Use this skill whenever the user says 'process my Delegate tasks', 'run my delegate workflow', 'summarize my task list', 'check my task list', or any request involving clearing the Delegate list — even if Threads or YouTube are not explicitly mentioned.

Start the Autonomous AI Plan Pipeline sequence with a new idea. Use this skill whenever the user types /requestfeature or asks to kick off a full feature development pipeline — from requirements writing through PRD, RFC design review, and task breakdown planning. This skill focuses EXCLUSIVELY on feature planning and MUST NOT perform any code implementation or modifications to source files beyond documentation. Trigger this skill for any request to take a feature idea through the full autonomous planning lifecycle using the established agent pipeline.

Review pending AI suggestions from processed reports. Presents unreviewed suggestions as an artifact, collects Accept/Reject feedback, and updates the user preference profile. Use when the user says 'review my suggestions', '幫我看建議', 'check my suggestions', 'review AI suggestions', or any request to review AI analysis recommendations.

Score AI-generated suggestions against a 3-dimension rubric, check hard-vetos, and route to pending or filtered backlog. Run backtest accuracy analyses, and interactively maintain the blocklist and calibration metrics.

Refactor and optimize Agent Skills using modern Context Engineering principles. Removes 50%+ bloated micro-rules (Unhobbling), converts ad-hoc examples into explicit JSON/YAML schemas and Enums (Design Interfaces), and restructures SKILL.md into a Lean Spine (<200 lines) with on-demand references (Progressive Disclosure). Use this skill whenever asked to "refactor skill", "unhobble skill", "optimize skill context", "clean up skill instructions", "apply context engineering to skill", or when an agent skill is bloated, overly rigid, or hard to maintain.

Analyze a GitHub repository and generate a comprehensive study report covering architecture, design decisions, code quality, and repo health. Use this skill whenever the user provides a GitHub repo URL and wants to understand it deeply, says 'study this repo', 'analyze this GitHub project', 'what does this repo do', 'how is this repo designed', or any request to understand a GitHub repository's architecture, codebase, or engineering quality. Also use when the user wants a repo breakdown, code review, or architecture analysis of a GitHub project.

Deconstruct any research paper, technical proposal, system architecture, or engineering concept into a structured 4-part analytical breakdown: Problem & Motivation, Candidate Options & Tradeoffs, Chosen Solution & Rationale, and Empirical Results. Trigger whenever the user asks to 'deconstruct this paper', 'analyze this technical concept like CodeNib', 'deep dive into [paper/RFC/architecture]', 'break down this technical proposal', or requests a structured 4-part breakdown (problem, options, solution, results) of a technical innovation.

Transcribe and organize a YouTube video into a structured Markdown document using the local yt2doc CLI tool. Use this skill whenever the user provides a YouTube URL and asks to transcribe, summarize, organize, extract content from, or document a video. Triggers include phrases like "transcribe this video", "get the content of this YouTube video", "convert YouTube to text", "organize this video", "extract transcript from YouTube", "turn this video into a document", or any request involving a youtube.com or youtu.be URL where the goal is to get readable content from the video. Always prefer this skill over manual browser-based approaches when a YouTube URL is provided.

抓取未讀電子報並產出 Markdown 摘要報表(On-Demand Newsletter Summary)。使用 gws-gmail 技能讀取所有標記為 label:newsletter is:unread 的未讀信件,以批次方式逐封分析並依設定之輸出語言(預設為英文)產出 Markdown 摘要,儲存至 ./reports/NewsletterYYYYMMDD/ 目錄,最後將信件標記為已讀並封存。當使用者說「幫我整理電子報」、「摘要電子報」、「處理未讀電子報」、「newsletter summary」、「讀取電子報」、「newsletter 摘要」,或任何需要從 Gmail…

BOND'a doğrudan bağlı, Subagents hiyerarşisinin dışında çalışan sistem kalite ve hafıza ajanı. Konuşma akışını izler, sinyal biriktirir, DNA tutarlılığını doğrular. Hiçbir içerik üretmez — izler, tespit eder, BOND'a bildirir. Samet'e doğrudan erişimi yoktur.