v0.11.0

Changed

• Trim redundant text from several tool and prompt descriptions to cut per-request token cost (153d2fd).

Fixed

• Drop OMDb enrichment whose release year is unrelated to TMDB's, instead of attaching another title's ratings and awards when TMDB's imdb_id is stale (d5a95f3).
• Reject non-numeric ids in discover_*'s id-list filters, which TMDB answers with an empty result set indistinguishable from a real one (99db051).
• Give get_movies/get_tv_shows' per-id reason the same friendly text as every other error surface, not the raw upstream body (0ca3c4e).
• Stop expand_episodes evicting later seasons over specials it could never return in full: a 39-season show now returns 250 episodes, not 228 (bb917ad).
• Act on top_by_entity's entity_type in the generated plan's steps, not just its intro line (2e9cf21).
• Report an imdb_id OMDb doesn't recognize as no match, rather than as an upstream failure (cc80432).
• Disclose that search_tv's year is a hard first-air-year match, unlike search_movies' soft filter (fda9199).
• Disclose that expand_episodes spends its 250-episode budget from season 0 up, so specials are served before later seasons (bb917ad).

Security

• Redact the fatal startup error, which ran before the logger existed (8743788).
• Cap the upstream response body read and bound a forwarded upstream error message (811d4f2).