v0.3.3

Accuracy + robustness release. Every per-tool claim re-verified against all 8 live Asta MCP tool schemas (zero drift); consolidates the 0.3.1→0.3.3 corrections on top of v0.3.0.

Correctness (verified vs live MCP schema)

• search_authors_by_name default fields="name" returns only name+authorId — workflows now request affiliations,paperCount,citationCount,hIndex,externalIds explicitly and prefer externalIds.ORCID for disambiguation.
• Documented all default limits: search_papers_by_relevance 50, snippet_search 20, get_citations 100, get_author_papers 1000 — with bounded examples to avoid context blowups.
• get_paper_batch.ids is a JSON array (not comma-separated, unlike snippet_search.paper_ids).
• venues matches Semantic Scholar's exact venue strings; get_author_papers accepts no venues; search_paper_by_title also supports publication_date_range.
• Date-filter null-publicationDate / Jan-1 fallback behavior documented.

Robustness / usability

• Task-specific field presets (metadata / search-ranking / DOI-export) + graceful degradation when pass-through fields are absent.
• Prerequisite Check: on missing tools, forbids raw HTTP / invented results and gives per-host registration hints.
• Output rules report a corpus-wide total only when the tool exposes one.

Maintenance / packaging

• Frontmatter nests non-standard keys under metadata (only name/description/license/metadata top-level) so it validates under Codex's stricter frontmatter rules.
• Added MAINTAINING.md: a pre-release checklist + pinned live-schema snapshot to catch future schema drift.