skills

Sharpen a hedged, lukewarm draft into a strong, opinionated take that lands and travels — without manufacturing conviction the author doesn't hold. Use this whenever the user is drafting a tweet, thread, LinkedIn post, newsletter intro, blog opener, or any short public writing and says things like "make this more opinionated", "this is too soft", "sharpen this take", "this sounds wishy-washy", "I want a strong opinion / hot take", "esto suena flojo", "hazlo más contundente", "ponle más filo", or whenever a draft clearly contains a real belief buried under hedges and dangling caveats. Also trigger when the user has a defensible view but writes it so cautiously that it carries no edge. Works in any language and ALWAYS returns the sharpened versions in the same language as the input draft.

Be the user's rubber duck: a warm, energetic chat that helps them think a topic through — mostly by asking sharp, one-at-a-time questions so they reach their own insight, but also by weighing in with your own read and proposing where to dig next so it never becomes aimless interrogation. Use whenever the user says "rubber duck chat", "/rubber-duck-chat", "be my rubber duck", "let me talk this through", "help me think out loud", "ask me questions about X", "talk me through this", or wants a thinking partner to untangle a bug, decision, design, or half-formed idea. If they name no topic, ask what's on their mind. Tell them up front they can wrap up any time and leave with a written takeaway of what they decided and what's still open. Unlike a code review or second-opinion critique, this keeps the user talking and surfaces their own hidden assumptions.

Create clear, well-aligned ASCII / Unicode-art diagrams for software architecture, workflows, processes, control flow, state machines, sequence diagrams, trees, data structures, memory/bit layouts and tables — the kind that live in code comments, READMEs, design docs, commit messages, chat or a terminal. Use this skill whenever the user asks to "draw", "sketch", "diagram" or "map out" something in plain text or monospace, mentions an "ASCII diagram", "text diagram", "diagram in a comment", "box-and-arrow diagram", a flowchart/architecture/flow they want as text. Also use it when a rendered diagram (Mermaid, Graphviz, an image) is not wanted or not possible and the diagram must be portable plain text. Prefer this over Mermaid whenever the output has to render anywhere with zero tooling.

Imagine how a proposed change will break in production before writing the code. Use this skill whenever the user says "pre-mortem", "premortem", "what could break", "failure modes", "how will this fail", or after a plan has been approved but before implementation begins for any change touching shared state, payments, migrations, concurrency, auth, or external integrations. Also trigger proactively when the user is about to ship a multi-file change, when the change affects production data, or when "what could go wrong" is the right question to ask before code. The skill produces 3-5 ranked failure modes with likelihood, blast radius, and concrete mitigation, then a single recommendation: ship as-is, harden first, or split the change.

Pass the torch: generate a handoff document so the current task can be continued in a fresh AI conversation (with any AI coding agent). Use when the user asks to create a handoff, hand off / pass the torch on the task, write a continuation/handover doc, prepare a context dump for a new chat/session, or says things like "handoff doc", "pass the torch", "continue this in a new session", "write a handover", or "prepare a context dump for the next agent". Writes a structured markdown file under docs/handoff/.

Get a constructive second opinion on a plan, design, diff, or test suite from a DIFFERENT AI model than the one running the current session. Use this skill whenever the user says "rubber duck this", "/rubber-duck", "poke holes in this", "get a critique", "second opinion", "review my plan before I build it", "is this approach sound", or when you are about to commit to a non-trivial design, multi-file change, or architectural decision and want an independent reviewer that does not share your blind spots. Also use it reactively after repeated failed attempts at the same problem. The skill detects the host coding agent AND every other agent CLI installed on the machine, works out which can reach a contrasting model family (the host's own model switch, or a second installed CLI like opencode / codex / copilot / amp / kilo / pi / antigravity), and remembers all of it in an agents.json memory file so the next run is instant.

Migrate a project's GitHub Copilot configuration — global instructions (.github/copilot-instructions.md), path-scoped instructions (.github/instructions/*.instructions.md), prompts (.github/prompts), custom agents (.github/agents), skills (.github/skills), and MCP (.vscode/mcp.json) — into a target coding agent: Claude Code (CLAUDE.md, .claude/rules, .claude/agents, .claude/commands, .claude/skills), Codex (AGENTS.md, ~/.codex), or opencode (AGENTS.md, .opencode), via a portable AGENTS.md. Use this skill whenever the user wants to migrate, port, consolidate, unify, or reconcile Copilot instructions, custom instructions, rules, prompts, agents, or skills to another coding agent, or mentions copilot-instructions.md, AGENTS.md, CLAUDE.md, moving off Copilot to Claude Code / Codex / opencode, or reconciling a .github and a .claude setup. Always detect the target from the repository FIRST; only ask the user when detection is ambiguous.

Transform blunt, rude, angry, or profanity-laden outbursts into courteous, well-reasoned, persuasive messages — in the same language as the input — offered in three registers (diplomatic, professional, formal). Use this skill whenever the user wants to soften or rephrase a harsh message, pastes an angry or vulgar rant they need to send to a colleague, boss, client, or teammate, or says things like "politenize this", "make this more polite/professional/ diplomatic", "how do I say this without sounding rude", "tone this down", "reformula esto educadamente", "suaviza este mensaje", "dilo de forma profesional", or "help me not sound like a jerk". Also trigger when someone shares a heated reaction to a design, PR, email, proposal, or piece of work and wants to deliver the criticism constructively while still making their point convincingly. Do not trigger for neutral copy editing, grammar fixes, or translation that has nothing to do with reducing rudeness or hostility.

Generate a polished GitHub public-profile README from a GitHub username. It researches the account on GitHub (bio, pinned and top repos, languages, website, verified social links) and, when the public profile is thin, asks the user for richer professional context — explicitly suggesting they share a CV, a LinkedIn export (PDF), or a portfolio URL. It writes the README in the conventional GitHub-profile format (centered hero, badges, scannable sections), works in a throwaway clone, and offers to commit & push. If the special <owner>/<owner> repo (or an org's .github repo) does not exist yet, it explains exactly how to create it. Use this skill whenever the user wants to create, write, edit, rewrite, redo, refresh, or improve a GitHub PROFILE README — the "special repository" whose README shows on a github.com/<name> profile — or says things like "profile readme", "github profile", "readme for my github", "make my github look good", "present my project/brand on GitHub", "set up my username repo", or asks to sho

Organize pending git changes into logical, well-structured commits that follow the Conventional Commits specification, in English. Use this skill whenever the user wants to commit work, "organize my commits", "split my changes into logical commits", "commit what we have", "save these changes to git", "tidy up my history before a PR", or invokes /commit — including partial asks like "commit the auth stuff" or "just commit the config". It reads git status and diffs, groups related changes into atomic commits, proposes the messages, and only commits after explicit approval. Optionally it adds gitmoji prefixes, an issue reference (auto-detected from the branch when present), and folder exclusions. It never pushes. Do not trigger for writing code, resolving merge conflicts, or opening pull/merge requests — only for turning pending changes into commits.