[jliounis]

Adds parallel-search-evaluation, an external onboarding skill based on the public search evaluation guide. It guides agents through product selection, representative gold sets, controlled single-hop/multi-hop comparisons, pilot inspection, grading, failure diagnosis, and end-to-end cost/latency reporting.

The bundled artifact contract defines dataset isolation, run manifests, trial traces, and reporting fields. README examples and a project discovery link expose the skill; the existing catalog and archive builders pick it up automatically. Current API details and pricing remain linked to public documentation.

Validation:

• All 42 repository tests pass with the released Parallel CLI installed.
• Markdown lint, skill frontmatter validation, and TypeScript type checking pass.
• Catalog and archive builds pass; verified the new catalog entry and both files in its archive.
• A fresh-agent onboarding trial produced a concrete plan with mode comparisons, three repetitions, gold isolation, grading and failure policies, and complete cost accounting. No paid evaluation calls were made.

Closes GTM-965.