Skip to content

Add end-to-end-doctesting skill#367

Open
JoshKappler wants to merge 2 commits into
mainfrom
add-end-to-end-doctesting-skill
Open

Add end-to-end-doctesting skill#367
JoshKappler wants to merge 2 commits into
mainfrom
add-end-to-end-doctesting-skill

Conversation

@JoshKappler

@JoshKappler JoshKappler commented Jul 16, 2026

Copy link
Copy Markdown
Contributor

Adds an agent skill that smoketests the setup docs by being the user: point it at a docs branch or PR and it sets up GT from scratch in a throwaway project, following the docs exactly as written, and reports where they break. Most first-time users now point an agent at the docs and say "install this" — an agent following the docs verbatim is the representative first-run experience, so where it gets stuck, users get stuck.

Usage:

/end-to-end-doctesting 3427               # docs on gt-cloud PR 3427 (resolves the content submodule SHA)
/end-to-end-doctesting content#354 next   # just the Next.js quickstart on a content PR
/end-to-end-doctesting main all           # every quickstart on the live docs

A single flow runs 10–30 min; with no scope it sweeps every actionable instruction in the tree. Output is a short glanceable verdict plus a findings report where each finding carries a concrete suggested edit against the exact file and line. It never posts review comments or pushes anything on its own — that only happens on an explicit "file it".

Trial runs against the docs-refactor branch (gt-cloud#3427) surfaced 7 real issues in the Next quickstart alone — e.g. the root-layout snippet silently drops globals.css/fonts/metadata, and the finished app ships no <html lang>.

Layout: the skill lives in .agents/skills/ next to docs-skill; .claude/skills/end-to-end-doctesting is a symlink so Claude Code picks it up, and agents/openai.yaml carries the metadata for OpenAI-style agents.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant