Tech pack QA agent — review every tech pack, BOM and costing in a folder

What you get

A drop-in agent for Claude Code (and the same procedure as a skill for Claude.ai, ChatGPT and Gemini) that walks a folder of tech-pack files — tech packs, BOMs and costing sheets as Markdown, CSV, text or PDF-as-text — reviews each against fixed checks (16 pre-production checks for tech packs, 8 BOM checks, 6 costing checks), writes one QA-<file>.md report per file and a QA-SUMMARY.md across files, and never modifies a source file. Findings come back as page/section/row → problem → fix → severity; unreadable files are asked about, not guessed. Runs in Claude Code, Claude, ChatGPT and Gemini.

Use it for

• Say "review the tech packs in ./packs" and get a QA report next to every file plus a summary
• Catch cross-file mismatches: BOM rows vs tech pack callouts, costing consumption vs BOM
• Redo one report after a pack is fixed, without re-running the folder
• Turn a style's findings into the factory email: do not start until / please note / please confirm

What's inside

• platforms/claude-code/.claude/agents/tech-pack-qa.md — the Claude Code agent: system prompt with procedure, checks, report and summary templates
• SKILL.md — how the QA pass works and the report format (also the Claude Code skill)
• PROMPTS.md — Start · Review one file · Summary across files · Factory email
• templates/QA-report.md — report skeleton with the tech pack, BOM and costing checklists
• examples/QA-example.md — a report and a summary written for a fictional field jacket folder
• platforms/chatgpt-instructions.md — Custom GPT / Project instructions
• platforms/gemini-gem.md — Gem instructions

Quick start

1. Unzip and open README.md.
2. Claude Code: copy platforms/claude-code/.claude/ into the folder that holds your tech packs, run claude, say "review the tech packs in ./packs". Claude.ai: add the files and SKILL.md to a Project and paste the Start prompt. ChatGPT / Gemini: paste the platform instructions and upload the files.
3. Read QA-SUMMARY.md first, then the QA-<file>.md reports; export scanned PDFs to text if the agent asks.

Preview — one prompt from the kit

Review only {{file}} ({{stage}} stage) with the tech pack QA checks. Detect its type (tech pack / BOM / costing) from its content and tell me which checks you applied. Return the QA report: header (file, type, style, stage, readability), verdict with counts, findings table "page/section/row → problem → fix → severity" most severe first, the ticked checklist ([x] pass · [ ] fail · [?] cannot verify, page or row on every fail), questions — one fact each. Do not change the file. Do not propose values in fixes.