Cloud & DevOps Workflows

Proactively catch common AWS infrastructure mistakes before they happen.

Contribute a new lesson to the AWS Learnings library (the llms.txt-format library at github.com/jbdamask/aws-learnings-library). Use this whenever you've just debugged, fixed, or discovered a non-obvious AWS gotcha — CloudFormation, CDK, Lambda, API Gateway, IAM, S3, CloudFront, EC2, EventBridge, SQS, Secrets Manager, SSM — and want to capture it for next time. Triggers on phrases like 'add an AWS learning', 'save this AWS lesson', 'capture this gotcha', 'add this to the learnings library', or after solving a tricky AWS deployment problem worth remembering. This skill WRITES files (a new lesson file + an updated index) and prepares a pull request; it is the write companion to the read-only 'aws-learnings' skill.

Create a token-secured REST API endpoint backed by AWS Lambda and DynamoDB with full CRUD and paginated list support. Deploys via CloudFormation. Use when user wants to quickly spin up an API endpoint, create a simple REST API on AWS, build a quick data store with an API, set up a Lambda+DynamoDB endpoint, or needs a fast way to log and retrieve structured data. Triggers on "create an endpoint", "quick API", "Lambda endpoint", "DynamoDB API", "CRUD endpoint", "log data to AWS".

Audit a codebase's test suite for tests that pass for no good reason — no assertions, tautologies, assertions on mocks only, swallowed exceptions, unawaited async assertions, permanently skipped tests, and other patterns that stay green no matter how broken the code is. Produces a red/yellow/green report. Use when the user says "bullshit tests", "find fake tests", "audit my test suite", "which tests are useless", "are these tests actually testing anything", "check test quality", "find tests that always pass", or asks whether green CI means anything.

Bulk-clone a set of git repositories into a single root directory using the bundled clone-repos.sh script. Use this whenever the user wants to clone several repos at once, pull down all the repos for a project or engineering estate, lay multiple repositories out side-by-side under one folder, or set up a local working tree of many repos (e.g. as the first step before reading, documenting, or doing an architecture review across a codebase). Trigger on phrases like 'clone these repos', 'clone all the repos', 'bulk clone', 'clone a bunch of repositories into a folder', or when the user hands over a list of git URLs and a destination directory.

Generate a succinct, factual, up-to-date company research report ('Company Short') in a fixed Markdown format. Use whenever the user asks to research a company, get a company overview/profile/brief, asks 'what does do', wants competitive or market intel on a business, or invokes 'company shorts'/'company short' with a company name or URL. Covers business model, products, financials, partnerships, market position, customers, and recent news.

Lints writing for AI slop tells, including overused words, cringe phrases, structural patterns, and bro-speak declaratives that are fingerprints of LLM-generated text. Use this skill any time the user asks to check, clean, lint, or review writing for AI tells, AI slop, cringe phrases, or generic LLM-sounding language. Also use it when the user says things like "does this sound AI-written?", "make this sound more human", "remove the AI vibes", "de-slop this", or "clean up my draft". If the user shares a piece of writing and expresses any concern about it sounding robotic, generic, or artificial, use this skill.

Writes documents in clear, human-sounding prose that avoids AI slop — the overused words, cringe phrases, structural tells, and bro-speak that fingerprint LLM-generated text. Use this skill whenever the user asks you to write, draft, compose, or put together a document of any kind: a blog post, essay, article, README, email, memo, announcement, newsletter, landing-page or marketing copy, report, proposal, cover letter, or similar prose. Use it by default for these writing tasks, not only when the user mentions sounding human or AI slop — though also use it when they say things like "write this without the AI vibes", "draft something that doesn't sound like ChatGPT", "make it sound human", or "no AI slop".

Evidence-based root cause diagnosis using a branching five-whys investigation. Produces a likelihood-ranked list of candidate root causes — diagnosis only, never a fix. Use ONLY when the user explicitly asks to diagnose a problem or find a root cause, with phrases like "find the root cause", "diagnose this", "do a five whys check", "root cause analysis", "why is this happening", or "run a five whys". Do NOT trigger on generic mentions of errors, bugs, or fix requests.

Play Napalm Death's 'You Suffer' — the shortest song ever recorded (1.316 seconds of grindcore). Use when the user says 'gilfoyle', 'you suffer', 'napalm death', 'play the song', or wants a moment of sonic brutality.

Final-phase skill that stages beads, commits, pushes to the remote branch, and labels the GitHub issue as "Ready". Invoked by issue-to-beads as the last step after all beads have been filed, reviewed, and synced. Can also be used standalone when you have synced beads on an issue branch that need to be committed and pushed.

Help users evaluate whether an MCP (Model Context Protocol) tool is safe and right for them before installing it. Use when a user wants to check, scan, review, or assess a third-party MCP server from GitHub, npm, or any repository URL. Provides plain-language assessments that anyone can understand, with technical details available on request.

Externalize detailed memory to Memory Bank before conversation summarization to prevent context loss. Triggers: CP, checkpoint, save, 存檔, 記一下, 保存, sync memory, dump, 先記著.

Spawn an Opus sub-agent in extended-thinking max-effort mode to debate, critique, and pressure-test the primary agent's current thinking through an unbounded multi-turn dialogue, terminating when the agents either reach genuine consensus or cleanly agree to disagree. Supports escalating clarifying questions to the human user mid-debate, and supports the user sending the debate back for another round (jury-room rule). Use when the user invokes "phone a friend", asks for a "second opinion", "sanity check", "debate this with another model", or otherwise signals they want rigorous adversarial review of an approach, plan, design, diagnosis, or decision before proceeding.

Plan-only mode that replicates Claude Code's built-in plan mode workflow but stops after the user approves the plan — no auto-execution. Use when the user wants to plan without executing, says "plan only", "just plan", "plan but don't execute", or when another skill needs planning without triggering auto-execution. Writes the plan to .claude/plans/ .md.

Create and configure Claude Code plugin marketplaces hosted on GitHub. Use this skill when users want to set up a new plugin marketplace repository, add plugins to an existing marketplace, configure team distribution settings, or troubleshoot marketplace issues. Triggers include "create a plugin marketplace", "set up a marketplace on GitHub", "add plugins to my marketplace", "configure marketplace for my team", or "distribute Claude Code plugins".

Security audit for GitHub repositories before installation. Use when user wants to check if a repo/app is safe to install, review install scripts for malicious code, verify an open source project isn't collecting data, or audit dependencies for suspicious packages. Triggers on phrases like "is this safe to install", "check this repo", "review this script", "audit this code", "is this sketchy".

Produces a concise, plain-language write-up of a technical subject that a smart but non-technical reader (like a manager) can fully understand — no jargon, but not dumbed down — and delivers it as a great-looking styled web page (an Artifact) by default, or as plain markdown if the user prefers. SEFAM means "Simple Enough For A Manager". Works on either a technical CHANGE (a PR, diff, commit, bug fix, GitHub issue) or a technical CONCEPT in a repo (a function, a process/workflow, an architecture, a system, a service, a design decision). Use when the user says "sefam", "SEFAM", "explain it like I'm not an engineer", or asks to explain how some function/process/architecture/system works for a non-technical audience.

Download transcripts (auto-captions) for videos from a YouTube channel within a date range, as timestamped markdown linked back to the source video. Use when the user wants to grab transcripts/subtitles/captions from a channel, e.g. "download transcripts for all AI Daily Brief videos from July 2026", "get the captions from this channel for last month", "pull transcripts between two dates". Requires yt-dlp and ffmpeg installed.