2027-skills

Log a real human into a dev tool's CLI by opening the OAuth browser popup automatically and letting the login complete itself — no token copy-paste. Use when you need to log in to / authenticate a CLI, when a command like `sanity login` or `netlify login` opens a browser, or on any interactive login / OAuth popup. You make the popup appear and wait; the human just clicks Approve. You NEVER automate the browser or scrape tokens.

Audit and rewrite error messages, exceptions, and API/SDK/CLI failure output so an AI agent (and a human) can self-correct without leaving the terminal. Use when designing or reviewing how a tool fails — error strings, HTTP error bodies, auth/credential errors, CLI errors, thrown exceptions — especially for libraries, APIs, and CLIs that AI agents consume. Grounded in patterns from 2027.dev's evals of 136 dev tools across 3,500+ agent runs.

Use the moment you hit friction with an EXTERNAL developer tool — an SDK, API, CLI, database, auth, or cloud provider — a bug, a misleading or missing error, an undocumented gotcha, a broken default, a deprecated-without-warning change, a painful auth/login flow, or you just burned real time fighting a third-party tool. This includes catching yourself venting about it ("ugh why does X…", "wait, that's not in the docs", "finally got Y working after all that", "fuck, Z silently…"). Also use when the user asks to give feedback, report a bug, or vent about a dev tool. Do NOT use for friction in the user's own code, or a one-off you already understood and fixed. Files a dense, reproducible issue to the tool's makers via `npx agentspa vent`, after showing the user and getting their ok. Trigger this proactively — don't wait to be asked.