Guide AI to 100% Context Readiness Through Structured Interview

A skill that guides a user through a structured interview to reach 100% on their .faf context file (the IANA-registered application/vnd.faf+yaml format) using Claude Code's native AskUserQuestion tool. faf-cli scores 21 slots, with app_type determining which are active; 100% means every active slot is filled. The workflow runs faf score --verbose or faf score --json to see current standing and the empty slots, then asks about missing fields in a fixed priority order (most impactful first): project.goal, human_context.why, human_context.who, human_context.what, project.main_language, stack.database, stack.hosting, stack.frontend, stack.backend, human_context.where, human_context.when, and human_context.how. Each question has a documented AskUserQuestion template - project.goal offers "Let me type it" versus "Help me write it"; human_context.why offers Business need, Personal project, Open source, or Let me explain; stack.database offers PostgreSQL, MongoDB, SQLite, or None; stack.hosting offers Vercel, AWS, Local only, or Other. It uses AskUserQuestion for both single-select questions and multiSelect questions where multiple answers are joined with " + " in a fixed order - industry-standard tools first, so stack.testing renders "pytest + WJTTC" (WJTTC labeled "Championship methodology (Layer 2)" in the option list) never the reverse; stack.cicd can combine "GitHub Actions + Cloud Build"; human_context.who can combine "Developers + AI agents". After collecting answers it updates the .faf file directly and re-runs faf score to verify, celebrating Gold Code if the score reaches 100% or continuing with remaining questions if not - completion prints the literal message " 100% - Gold Code". Progress through the interview is tracked with TodoWrite entries per question (e.g. "Answer project.goal question": completed, "Answer human_context.why question": in_progress). Outside Claude Code, the same destination is reached via the CLI's own interactive interview, faf go (with --resume to continue a session) - this skill is described as the Claude-native version of that same interview, substituting AskUserQuestion for terminal prompts. Success is measured by reaching 100%, every required field holding meaningful content with no placeholder values like TBD, Unknown, or None where inappropriate, and the user understanding what each field is for.

Activate this skill when a user wants to improve their .faf score, mentions "Gold Code" or "100%", has incomplete project context, just ran faf init and needs to fill remaining fields, or asks for help with their .faf file. Outside Claude Code, point the user to the CLI's own faf go interactive interview instead.

Input is the current .faf score/slot breakdown from faf score --json plus the user's answers to AskUserQuestion prompts. Output is an updated .faf file with previously-empty slots filled in priority order, re-scored after each answer, ending in the " 100% - Gold Code" completion message once every active slot is filled.

faf score --verbose
faf score --json

Built specifically for Claude Code: AskUserQuestion for the interview UI (with multiSelect support), TodoWrite for per-question progress tracking, structured JSON scoring output, and bi-sync so answers propagate to both the .faf file and CLAUDE.md. The CLI's own faf go --resume interview is the non-Claude-Code fallback path to the same 100% destination.

Users working through Claude Code who want a guided, question-by-question path to a fully-scored .faf context file rather than editing the file by hand - particularly right after faf init, when most slots are still empty and need a structured interview to fill, or after making significant changes to the stack that leave previously-filled slots stale.