Add category-monitor skill

[norightsforlis]

Summary

Adds the category-monitor skill: Weekly monitor of a category's aggregate health (estimated sales, products, catalog products, sellers, medal mix, monopolization), reporting what changed since last week.

Lives in skills/category-monitor/SKILL.md and is registered in the README skills table. Public behavior and user-visible workflow only — no internal service names, schema, or implementation details.

Public-Safety Checklist

• No credentials, tokens, keys, or secrets.
• No internal filesystem paths.
• No private Slack, Jira, Notion, admin, or service URLs.
• No private service names or implementation details.
• No customer data or private product data.
• Skill directory name matches SKILL.md frontmatter name.

🤖 Generated with Claude Code

[profunctor]

KISS review: the skill currently promises week-over-week monitoring without a usable baseline contract.

It says each run is its own session, later runs should read the previous recorded table, and the skill is read-only; in a fresh invocation the agent cannot recover that baseline, so it may fail or infer/fabricate changes. Please either make the previous report/table an explicit required input or remove the ongoing monitor/diff promise until storage/automation exists.

Also, it routes one-shot snapshots to opportunity-index, which is not present in this repository/catalog. Please replace that with an available skill or remove the reference.

[profunctor]

Review — category-monitor

Summary: A weekly, read-only monitor of one Mercado Livre (Brasil) category's aggregate health via JoomPulse (estimated sales, product/catalog counts, active sellers, medal distribution, monopolization), reporting era → agora diffs vs the previous run. First run is a snapshot.

Findings:

• major — Dangling reference to a non-existent opportunity-index skill (twice in the description, once in the body). No such skill exists on main. Either drop the references or land that skill first.
• minor — Baseline persistence is underspecified. The workflow says "read the previous run's recorded table" and "each run is its own session," but never says where that table is stored or how a fresh session retrieves it. As written, a clean session has no baseline source, so it may always behave as a first run. A sentence on how the prior table is supplied would fix this.
• nit — No downloadable export, unlike product-change-monitor / ml-product-analysis. Reasonable given the small aggregate table, but worth a deliberate note.

Consistency is strong: frontmatter shape, section ordering, the estimate disclaimer, README row, and folder-name == name all match existing skills. No secrets, internal paths, broken queries, or unsafe operations.