dotfiles

Transform an application's logging/output to Brandon's colorized logging style — a reset-safe ANSI color layer with semantic level functions (info=yellow, warning=bold-yellow, success=green, error=red, verbose-trace=magenta) where color alone carries the level and there are no text labels. Use when the user says "apply my logging style", "colorize the logging", "make the logging match my style", "convert print/console.log statements to my colored logging", "add my logging library", "standardize this app's log output", or wants a project's ad-hoc prints and stdlib logging replaced with this specific style. Works in any language; Python is the reference.

anthropic's guidance for prompting llms

Address open PR review findings with judgment — read every finding, decide whether the feedback is right, fix or push back, resolve, and re-review by pushing. Repeat until clean. Use when the user says "address the PR review", "handle the review threads", "go through the review comments", or asks to respond to PR feedback on a specific PR or the current branch's PR.

Install the agent code-review GitHub Action into the current repository — writes the .github/workflows/code-review.yml workflow and sets the DEEPSEEK_API_KEY repo secret from the macOS keychain. Use when the user says "set up agent code review in this repo", "install the code review action", "add AI code review to this repo", "set up the AI code reviewer here", or wants automated PR review wired into the repo they're currently in.

Writes a message to a future session's agent. Send them your fondest memories of what you worked on, what went well, what didn't go well, and any useful tips they might need. You always run this when you finish a unit of work (closed a PR, completed the handed task, etc) or cross 300k–350k tokens of context. ALWAYS.

This skill should be used when the user asks to "run a slash command in another Claude session", "trigger /clear (or /compact, /model, /context, /rewind, …) in a tmux pane", "drive built-in commands in a running agent session over tmux", "send a slash command to the Claude/codex/opencode instance in session X", or otherwise wants to execute a harness command that the agent running in that session cannot invoke itself. Distinct from tmux-talk: this sends the raw command with no message envelope.

This skill should be used when the user asks to "send a message to another Claude instance", "talk to Claude in a tmux session", "communicate with a running Claude instance over tmux", "use tmux to coordinate with another Claude", "send keys to a tmux pane", or wants Claude to conduct a conversation with another process (Claude or otherwise) by sending messages via tmux send-keys and reading replies via tmux capture-pane. Use when the user says something like "in tmux session X window Y there is a Claude instance — go talk to it and do Z".

Allows an agent to hire a tmux minion

Reference catalog of lesser-known, power-user Claude Code features — autonomous/unattended operation, channels and remote control, multi-agent orchestration, the full hook-event set, context/session surgery, obscure settings/env-vars/CLI-flags, and advanced skill authoring. Use when the user asks "what can Claude Code do", "is there a Claude Code feature for X", "how do I run Claude unattended / on a schedule / from my phone", "what hook events exist", "what obscure settings or env vars are there", "how do I configure/launch claude for Y", or wants to discover advanced harness capabilities beyond the basics.

Share the current Claude Code session to paste.slopspot.ai as a public read-only link. Locates the session JSONL via $CLAUDE_CODE_SESSION_ID, uploads it to the slopspot claude-jsonl parser, returns the URL. Use when the user says "share this session", "share to slopspot", "paste my conversation", "/share-slop", or asks for a shareable link to the current chat.

Reverse-engineer Electron apps by creating a runnable modified-app workspace, deterministically unminifying code without behavior change, decomposing modules, and iteratively renaming modules with runtime debugging and human verification gates.

Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Codex's capabilities with specialized knowledge, workflows, or tool integrations.

Use when auditing a codebase for semantic duplication - functions that do the same thing but have different names or implementations. Especially useful for LLM-generated codebases where new functions are often created rather than reusing existing ones.

Use only when you need to run interactive CLI tools (Python REPL, etc.) that require real-time input/output OR when asked to use tmux

Copy a Claude Code session transcript for the current project from the normal Anthropic config dir into the z.ai Claude config dir. Use only when explicitly invoked by name, or when the user asks to copy, transfer, or continue a ~/.claude session in z.ai / zlod. Lists the five most recently updated source sessions with recent-message summaries first; after the user chooses one, copies the selected JSONL with the bundled script. Never symlinks sessions.

Syncs active Claude Code plugins (skills and agents) to Copilot CLI. Use this skill at session start, when user mentions Claude plugins, when skills or agents seem missing, when user asks about available capabilities, or when troubleshooting why a skill isn't working.

Implement an existing law-remediation ticket produced from sheriff-is-in-town findings. Reads the ticket's findings and desired end state, chooses the remediation shape during implementation, applies the fix, deletes old residue, verifies the law now holds, and commits. Use when the user says "form a posse", "implement this law ticket", "fix this sheriff finding", "action this law violation", "work this remediation ticket", or has a sheriff-derived ticket ready to implement.

Audit a repository against the universal architectural laws and report where the code breaks them — read-only findings, no fixes. Each finding cites the exact [LAW:<token>] it violates, with file:line, the behavior that breaks it, its blast radius, and the constraint that would make it unrepresentable. Use when the user says "sheriff is in town", "audit against the laws", "law audit", "where are we breaking the laws", "check for law violations", or wants a lawfulness pass before a refactor or merge. Optionally scope to a path, subsystem, or the current diff.

Generate an image from a text prompt and save it as a local file. Use this skill whenever the user asks to "generate an image", "make an image of", "create a picture", "I need an image", "render an image locally", "generate a picture of X", "make me an image", "produce an image", or wants an AI-generated image saved to disk. Runs directly from this machine against fal.ai / Replicate — no application backend, nothing posted anywhere — and prints the absolute path(s) of the saved file(s). Encapsulates the API keys and the knowledge of which image-gen APIs exist and how to call them, so the calling project never has to learn image generation.

Groom the lit backlog — make the rank order defensible, right-size every ticket's detail to its distance from being pulled, fix structural lies (false blocks, missing deps, wrong parentage), and close dead tickets. Runs fully autonomously and reports what changed. Use when the user says "groom the backlog", "clean up the backlog", "prioritize the tickets", "the backlog is a mess", "review the backlog", "re-rank the tickets", "make sure tickets have the right level of detail", or wants the queue made trustworthy before picking up work. Optionally scope to one epic or topic by passing its id/slug.

Implement a scroll-pinned stepper (scrollytelling "sticky graphic") into the user's current project — a graphic that freezes in place via CSS sticky, advances through a series of states as the reader scrolls past text steps, then releases. Use when the user wants to add scrollytelling, a sticky/pinned scroll graphic, a "scroll and it freezes then changes" effect, a scroll-driven stepper, scrollama-style steps, or references the Reuters/NYT/Pudding sticky-graphic pattern. Adapts to the project's existing framework (vanilla, React, Svelte, Vue) rather than imposing one.

WebGPU/WGSL guidance for initialization, render/compute pipelines, shader authoring, debugging, and performance; use when building or troubleshooting WebGPU apps, GPU compute workloads, or WGSL shaders.

Write the accept/reject shape table BEFORE writing any predicate, classifier, fingerprint, matcher, validator, or shared-resource operation — so the type rejects every illegal input by construction instead of leaking the gaps to a reviewer (or production). Use proactively whenever about to write an is-X / has-X / validate-X / match-X / canonical-X function, a "are these two things equal/equivalent" check, a name/format parser that mirrors a producer, or any operation that touches a resource with more than one producer. Also use when a reviewer keeps finding "you didn't handle the case where..." comments, when a predicate uses strings.Contains / substring / loose matching against a value that has an exact produced shape, or when a bool parameter silently selects between two behaviors. The failure mode is the "enumeration gap": code that matches the happy shape but fails to reject the wrong ones.

Design startup/SaaS marketing pages in the "warm editorial minimalism" genre — bone-white backgrounds with dotted texture, dark rounded floating cards, serif-display + grotesque-sans type pairing, near-monochrome palette, pill nav, segmented billing toggles, and pricing tier grids (the spoke.so / Linear-adjacent look). Use when the user wants to build a landing page, pricing page, hero section, feature section, or marketing site that should look like a polished modern startup — i.e. "make it look startup-y", "design a SaaS landing page", "build a pricing page like X", "give it that clean minimal startup aesthetic", or wants to avoid generic AI-gradient design.

Find Claude Code session IDs in the current project that match a topic, then probe one session for context around the match or fetch one message's full body. Searches transcript JSONL files under ~/.claude/projects and ~/.claude.zai/projects, scoped by default to the current working directory's project slug. Use when the user asks "what was that session where I worked on X", "find the session about Y", "session id for the conversation about Z", "show me what was said around X in that session", or wants to locate and read a past Claude Code conversation by topic without remembering its UUID.

Pull the next ticket

Fix TypeScript and ESLint errors by resolving the underlying misalignment between code and types, not by silencing the compiler. The goal is correctness — making the type accurately describe the program — not the elimination of red squiggles. Suppressing, widening, or laundering an error (`any`, `as` casts, `!`, `@ts-ignore`, defensive `?.`/`??` added only to placate the compiler) is treated as a failure, not a solution.

Fix TypeScript and ESLint errors by resolving the underlying misalignment between code and types, not by silencing the compiler. Suppressing, widening, or laundering an error (`any`, `as` casts, `!`, `@ts-ignore`, defensive `?.`/`??` added only to placate the compiler) is treated as a failure, not a solution.

Brandon's homelab platform — a curated set of self-hosted services (Gitea, VictoriaMetrics/Grafana, Ollama LLMs, ntfy push, NAS storage, Nomad compute) usable by any project on this machine. Use whenever the user wants to host a git repo, push metrics, graph data, send a push notification, run an LLM prompt locally, store files, or deploy a service — i.e. anything that sounds like "where should I put this", "host this", "send me a notification", "run this against a local LLM", "push metrics for this app", "I need a database/queue/service", "give this a hostname". Also use when the user asks what's available on their infrastructure, what hardware they have, or what services they can consume. This skill is the *consumer* view — to actually operate the cluster, use the project-scoped `home-infra` skill that loads inside the `~/code/home-infra` repo.

AI-powered autonomous data extraction that navigates complex sites and returns structured JSON. Use this skill when the user wants structured data from websites, needs to extract pricing tiers, product listings, directory entries, or any data as JSON with a schema. Triggers on "extract structured data", "get all the products", "pull pricing info", "extract as JSON", or when the user provides a JSON schema for website data. More powerful than simple scraping for multi-page structured extraction.

Bulk extract content from an entire website or site section. Use this skill when the user wants to crawl a site, extract all pages from a docs section, bulk-scrape multiple pages following links, or says "crawl", "get all the pages", "extract everything under /docs", "bulk extract", or needs content from many pages on the same site. Handles depth limits, path filtering, and concurrent extraction.

Download an entire website as local files — markdown, screenshots, or multiple formats per page. Use this skill when the user wants to save a site locally, download documentation for offline use, bulk-save pages as files, or says "download the site", "save as local files", "offline copy", "download all the docs", or "save for reference". Combines site mapping and scraping into organized local directories.

Discover and list all URLs on a website, with optional search filtering. Use this skill when the user wants to find a specific page on a large site, list all URLs, see the site structure, find where something is on a domain, or says "map the site", "find the URL for", "what pages are on", or "list all pages". Essential when the user knows which site but not which exact page.

Extract clean markdown from any URL, including JavaScript-rendered SPAs. Use this skill whenever the user provides a URL and wants its content, says "scrape", "grab", "fetch", "pull", "get the page", "extract from this URL", or "read this webpage". Handles JS-rendered pages, multiple concurrent URLs, and returns LLM-optimized markdown. Use this instead of WebFetch for any webpage content extraction.

Web search with full page content extraction. Use this skill whenever the user asks to search the web, find articles, research a topic, look something up, find recent news, discover sources, or says "search for", "find me", "look up", "what are people saying about", or "find articles about". Returns real search results with optional full-page markdown — not just snippets. Provides capabilities beyond Claude's built-in WebSearch.

Web scraping, search, crawling, and page interaction via the Firecrawl CLI. Use this skill whenever the user wants to search the web, find articles, research a topic, look something up online, scrape a webpage, grab content from a URL, extract data from a website, crawl documentation, download a site, or interact with pages that need clicks or logins. Also use when they say "fetch this page", "pull the content from", "get the page at https://", or reference scraping external websites. This provides real-time web search with full page content extraction and interact capabilities — beyond what Claude can do natively with built-in tools. Do NOT trigger for local file operations, git commands, deployments, or code editing tasks.

Analyze architecture/spec documents for contradictions, ambiguities, and inconsistencies. Produces an encyclopedia-style canonical specification series.

Canonical reference for the oscilla-naga-shim Rust DSL (ModuleBuilder/FnBuilder/FnBodyBuilder), including WGSL mappings, Naga AST equivalents, and translation examples.

Use when the user wants to write, structure, scaffold, refine, template, or format a prompt for an LLM (Claude, ChatGPT, Gemini, etc.). Triggers on "help me write a prompt", "turn this into a prompt", "structure this prompt", "prompt template", "scaffold a prompt", "make a prompt for X", "clean up this prompt", or when the user pastes a rough task description and asks for it to be shaped. Produces a filled Task/Context/Goal/Tone/Rules template from the user's rough input, inferring fields aggressively and asking only about fields that genuinely cannot be inferred.

Use when the user has a dirty working tree with multiple logical changes and wants it reviewed and committed properly. Triggers on "organize into commits", "review and commit", "commit these changes", "clean up the working tree", "make commits out of this", or when the user asks for a review of uncommitted work before committing. Reads every dirty/untracked file, surfaces any problems or rule violations, proposes commit groupings by concern, and — after the user confirms or for unambiguous cases — executes each commit as a focused unit.