← ClaudeAtlas

walk-the-recipe-as-target-userlisted

Validation discipline for any "documented workflow" (README reproduction recipe, getting-started guide, install script, onboarding doc, deploy procedure). Simulate the ACTUAL target user's environment — different auth path, different SDK credentials, different OS, different tooling — and walk every documented step as that user would. Each step that breaks for the target user is real friction. Each step that silently assumes "you already have X set up" is an undocumented precondition. Use when: (1) the workflow was written by a developer whose machine has every privilege/credential/tool pre-configured and you're worried real users don't, (2) "first-touch hardening" or "on-ramp polish" or any phase whose stated purpose is making the workflow reachable for non-developers, (3) you've just changed the underlying CLI/API surface and the documented workflow predates the change, (4) a literal-spec adherence move (do the checklist items the doc names) would miss the friction the doc was MEANT to address. Catches the s
MrBinnacle/skills · ★ 0 · AI & Automation · score 65
Install: claude install-skill MrBinnacle/skills
# Walk the Recipe as the Target User ## Problem A documented workflow — README reproduction recipe, onboarding script, install procedure, deploy guide — works perfectly for the developer who wrote it. They have all the credentials, all the SDKs, all the tooling already installed. They wrote the recipe by literally running the steps and confirming each one. They aren't lying about it working. But they wrote it on a privileged-environment machine. Their `ANTHROPIC_API_KEY` is set. Their `KUBECONFIG` points to the cluster. Their `gcloud auth` is current. Their AWS SSO session is fresh. Their `node_modules` is hydrated. Their docker daemon is running. A real user — a T1 reader, a new hire, a contractor, an external evaluator, an open-source maintainer reviewing a PR — has a DIFFERENT environment. Their version of "I cloned and tried to follow the README" is structurally different from the author's "I followed my own README." The differences are invisible to the author because the author cannot reproduce their own absence-of-a-credential. The documented recipe stays accurate as far as the author can see, and the recipe stays broken for the real user. The literal-spec response — "go through the 6-item hardening checklist the planning doc names" — does not catch this failure mode. The 6 items address the author's THEN-CURRENT best guess of friction. Reality friction may have moved, may include items the author never considered (because their own privilege blinded them), or may