Promote mlops from Snowflake-Solutions

[jdanielmyers]

Summary

Promotes the mlops skill from the internal Snowflake-Solutions/cortex-code-skills repo to public Labs, rewritten for an external builder audience (developers, data engineers).

Run through the v1.1.0 audit pipeline:

• Holistic LLM rewrite for builder audience
• All internal Snowflake references substituted with public equivalents (e.g., Snowhouse -> ACCOUNT_USAGE)
• audience.no_internal_refs check passes
• Sub-skills folded into <sub>/INSTRUCTIONS.md per the umbrella precedent (manage-zerocopy-sapbdc)

Test plan

• Audit blocking_failures == 0
• audience.no_internal_refs passes
• Frontmatter matches Labs conventions (snowflake-docs / manage-zerocopy-sapbdc precedents)
• LICENSE present (Snowflake)
• No Snowflake-Solutions repo path leakage
• No nested SKILL.md in sub-skill dirs (all converted to INSTRUCTIONS.md)

[jdanielmyers]

Audit Report: mlops

Re-audit (post-v1.1.1 + critical fixes). Original report superseded.

Summary

Category	🔴	🟡	🟢
Frontmatter	0	0	0
Structure	0	0	0
Workflow	0	0	1

Critical 🔴

None — all critical findings from the original audit have been addressed.

Warnings 🟡

None.

Suggestions 🟢

1. Strong "scope split" table clearly delineating ownership between mlops and machine-learning. The "Explain before asking" directive and "per-message intent re-evaluation" rule are good patterns for preventing the agent from assuming user knowledge or losing track of shifting intent. "Red Flags" section is thorough.

Verdict

Ready to merge — Clean router with well-defined scope boundaries, explicit promotion-pattern prerequisite, per-turn intent re-evaluation, and refused rationalizations pattern.