refactor(query): extract SOCIAL_NOISE and VIRAL_NOISE shared sets

[iliaal]

Summary

Six adapters defined near-identical noise frozensets inline inside their _extract_core_subject wrapper. The architecture review surfaced this in the _extract_core_subject consolidation finding (Architecture F5). Move the shared sets to lib/query.py as SOCIAL_NOISE and VIRAL_NOISE; have the adapters reference them.

Mapping

• SOCIAL_NOISE (18 words): Bluesky, Threads, Truth Social. Short-form micro-social platforms where research/meta words rarely appear in post bodies.
• VIRAL_NOISE (25 words = SOCIAL_NOISE + 7 extras): TikTok, Instagram, Pinterest. Adds killer, the prompt-meta cluster, and the methodology cluster.
• YouTube extends VIRAL_NOISE with temporal/meta tokens (months, recent year strings, etc.) that the planner emits but YouTube titles don't carry. Now composed as VIRAL_NOISE | _YT_EXTRA.

What's NOT in scope

• Wrappers stay. The wrappers themselves are 3-line shims that document each adapter's noise choice; removing them would force every callsite to know the right noise set. Future cleanup if desired.
• Polymarket's _extract_core_subject is a custom prefix-stripper that doesn't go through query.extract_core_subject at all. Out of scope.
• Reddit uses the module-level default NOISE_WORDS. Out of scope.

Test plan

• New TestSharedAdapterNoiseSets pins SOCIAL_NOISE membership (18 words), VIRAL_NOISE = SOCIAL_NOISE + 7 specific extras, and verifies extract_core_subject behavior with both sets.
• Set arithmetic verified: old _YT_NOISE (52 items) = new VIRAL_NOISE | _YT_EXTRA (25 + 27 = 52). Zero behavior change for YouTube.
• Existing adapter callsites unchanged; the wrappers continue to return the same value.
• Full suite: pytest tests/ --ignore=tests/test_exa_search.py: 1556 passed, 4 skipped.

[greptile-apps]

Greptile Summary

This PR extracts six near-identical inline noise frozensets from adapter _extract_core_subject wrappers into two shared module-level constants — SOCIAL_NOISE (18 words, used by Bluesky/Threads/Truth Social) and VIRAL_NOISE (SOCIAL_NOISE + 7 extras, used by TikTok/Instagram/Pinterest) — in lib/query.py, and composes YouTube's set as VIRAL_NOISE | _YT_EXTRA. The arithmetic is verified: all seven adapters produce frozensets identical in content to their pre-refactor inline definitions.

• lib/query.py: Two new public constants (SOCIAL_NOISE, VIRAL_NOISE) added; no changes to any existing function or constant.
• Six adapters (bluesky, threads, truthsocial, tiktok, instagram, pinterest): inline frozenset definitions replaced by a single import; youtube_yt replaces its full inline set with VIRAL_NOISE | _YT_EXTRA.
• tests/test_query.py: New TestSharedAdapterNoiseSets class pins exact membership of both constants and exercises extract_core_subject with each.

Confidence Score: 5/5

Pure mechanical refactoring — every adapter's effective noise set is byte-for-byte identical before and after, verified by tests and manual count. Safe to merge.

All seven adapters produce frozensets identical in content to their pre-refactor inline definitions; the new shared constants are pinned by membership-equality tests; no logic, API surface, or call sites were modified.

No files require special attention; the only open nit is the inline frozenset in TestCustomNoise.test_custom_noise_keeps_tips that duplicates VIRAL_NOISE.

Important Files Changed

Filename	Overview
skills/last30days/scripts/lib/query.py	Added SOCIAL_NOISE (18 words) and VIRAL_NOISE (SOCIAL_NOISE + 7 extras) as module-level frozenset constants; no changes to existing logic.
skills/last30days/scripts/lib/bluesky.py	Replaced inline _BSKY_NOISE (18 words) with SOCIAL_NOISE import; identical frozenset content, zero behavioral change.
skills/last30days/scripts/lib/instagram.py	Replaced inline _INSTAGRAM_NOISE (25 words) with VIRAL_NOISE import; identical frozenset content.
skills/last30days/scripts/lib/youtube_yt.py	Replaced inline _YT_NOISE (52 items) with VIRAL_NOISE
tests/test_query.py	Added TestSharedAdapterNoiseSets with 4 tests pinning SOCIAL_NOISE membership, VIRAL_NOISE superset relationship, and extract_core_subject behavior; existing TestCustomNoise still carries an inline frozenset identical to VIRAL_NOISE.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    SN["SOCIAL_NOISE\n18 words\nBluesky · Threads · Truth Social"]
    VN["VIRAL_NOISE\nSOCIAL_NOISE + 7 extras\nTikTok · Instagram · Pinterest"]
    YT["VIRAL_NOISE | _YT_EXTRA\n25 + 27 = 52 words\nYouTube"]

    SN -->|"+ killer, prompt*,\nmethods, strategies,\napproaches"| VN
    VN -->|"+ temporal/meta tokens\n(months, years, etc.)"| YT

    subgraph query.py
        SN
        VN
    end
    subgraph youtube_yt.py
        YT
    end

Loading

_{Reviews (2): Last reviewed commit: "refactor(query): extract SOCIAL_NOISE an..." | Re-trigger Greptile}

[tmchow]

@iliaal fix conflicts please

[iliaal]

@tmchow rebased on top of main. Conflicts (all in the test files touched by the conftest.py centralization that merged earlier today) are resolved.