You are helping the user quickly define what to extract from a website. Download 1 detail page, discover fields, and iterate on the schema in the terminal until approved.

Read ${CLAUDE_SKILL_DIR}/../scrape/references/python-environments.md.

The output is a draft spec folder with an approved schema and values (no stored pages — those are Stage 2's job). It can be expanded with /scrape-spec (more pages, variant comparison, navigation).

Hard constraints — never violate these:

You MUST NOT fetch, read, grep, or parse any HTML file yourself. Page download is handled by the /scrape-explore-site subagent; field discovery is handled by the /scrape-analyze-page subagent. The main agent only orchestrates and consumes their outputs.
This includes raw.html, rendered.html, *.cleaned.html, and any other saved page HTML. You may check whether a file exists, but you MUST NOT open it, print it, grep it, parse it, or run extraction scripts against it from the main agent.
You MUST invoke /scrape-analyze-page as a subagent before building any schema. Building a schema from raw HTML without first running that subagent is a critical error.
While waiting for /scrape-analyze-page, do NOT read HTML files, run clean_html.py, extract_metadata.py, or any inline parser code, even if the agent response suggests continuing unrelated work. Wait for the subagent to complete, then read only its saved JSON output.

scrape-define

Parse intent