The dedup hot path recomputed normalize_text() 4 times per comparison
and recomputed item_text() on every inner-loop iteration. Pre-computing
n-gram sets and token sets into a _PreparedText cache cuts dedup time
by 6x (2.16s to 0.39s on 300 unique items).
Bird handle searches spawned one Node process per handle sequentially.
Now uses ThreadPoolExecutor so N handles run concurrently. Same pattern
applied to YouTube comment enrichment (was serial, Reddit was already
parallel) and the retry-thin-sources phase in the pipeline.
Clustering now pre-computes candidate text and uses prepared_similarity
for the O(n^2) grouping and MMR representative selection loops.
Minor: _is_wsl() cached with lru_cache, Bundle.add_items() uses
extend() instead of list concatenation.
End-to-end: 5.2s -> 3.7s (29% faster) on a typical 4-source query.
On WSL2, native Linux Firefox typically has no x.com cookies since users
browse in Windows. Chromium browsers (Edge, Chrome, Brave) encrypt cookies
with DPAPI/app-bound encryption, making them inaccessible without admin
privileges. Windows Firefox stores cookies unencrypted in SQLite, readable
directly through the /mnt/c mount.
The cookie extractor now detects WSL2 via /proc/version, locates Windows
Firefox profiles under /mnt/c/Users/*/AppData/Roaming/Mozilla/Firefox,
and falls back to them when Linux Firefox yields no results. Reports
source as "firefox-wsl" to distinguish from native.
Also fixes profile resolution priority: Install* sections (Firefox >= 67)
now take precedence over the legacy Default=1 flag, which could select a
stale profile on multi-profile installations.
When Bird's JSON response is a raw array instead of an object,
json.loads returns a list. All callers use .get('items') which raises
AttributeError on lists. Wrap list responses in {"items": parsed} so
callers always receive a dict.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat: INCLUDE_SOURCES config + TikTok/Instagram opt-in in NUX
- INCLUDE_SOURCES=tiktok,instagram in .env forces sources on for all
query types, bypassing the tier system
- NUX shows opt-in modal after ScrapeCreators key is saved: "Also
search TikTok and Instagram?" with honest call-usage warning
- Tier system preserved as default — override only when INCLUDE_SOURCES set
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: neutral call-usage copy — works for free and paid tiers
---------
Co-authored-by: Matt Van Horn <mvanhorn@MacBook-Pro.local>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add extract_transcript_highlights() that scores sentences by specificity
(numbers, proper nouns, topic relevance) and filters YouTube filler
(subscribe, welcome back, etc). Top 5 highlights shown as structured
bullets in compact output. Full transcript moved to collapsible <details>
block so the LLM reads highlights first, full text on demand.
SKILL.md updated to instruct the judge agent to quote highlights
directly in synthesis, same as Reddit top comments.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
TRANSCRIPT_MAX_WORDS raised from 500 to 5000 so the LLM gets the full
content of most videos (up to ~25 minutes). Removed the second 200-char
truncation in render.py that was reducing transcripts to a single sentence
before the judge agent ever saw them.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When Cloudflare blocks requests to bsky.social or public.api.bsky.app
with a 403, the error was swallowed by a generic except clause and
reported as "Bluesky auth failed" - misleading users into thinking
their credentials were wrong.
Now _create_session() preserves the specific error in _session_error,
and search_bluesky() surfaces it. Cloudflare 403s get a clear message
about network-level blocks. Actual 401s say "Invalid credentials".
Closes#69
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Updated frontmatter description to be more search-friendly for ClawHub.
Added 12 new tags: deep-research, twitter, bluesky, recency, news,
citations, multi-source, social-media, analysis, web-search, ai-skill,
clawhub. Also added 11 GitHub repo topics.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The short alias /last30 only works on some platforms. Claude Code requires
the full /last30days name, so the follow-up suggestions after comparison
research were producing "Unknown skill: last30" errors.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Map prompt-oriented product searches and animation-oriented build searches away from the breaking-news default so source tiering and tiebreakers align with the benchmark topics.
Validation: uv run python -m unittest tests.test_query_type
Isolate eval subprocesses from local yt-dlp config and fix nDCG normalization against the judged pool.
Validation: uv run python -m unittest tests.test_evaluate_search_quality
This workspace uses GOOGLE_API_KEY as the canonical Google credential. Accept it ahead of the Gemini-specific aliases so the local evaluation harness can run without a separate GEMINI_API_KEY export.
Validation: uv run python -m unittest tests.test_env_project tests.test_evaluate_search_quality and a one-shot keychain-backed resolution check.
Add an optional local evaluator that compares a baseline revision against a candidate checkout, computes deterministic stability metrics, and can call Gemini for judged ranking metrics when configured.
The harness isolates child runs with a temporary HOME and a node-free PATH so historical revisions cannot trigger Bird browser-cookie auth during evaluation.
Validation: uv run python -m unittest and local smoke/full deterministic eval runs.
Score against original user intent on Reddit, remove the artificial low-end relevance floor, and make Polymarket semantics dominate generic market quality signals.
Also apply the relevance filter to Polymarket and update the affected cross-source tests.
Validation: uv run python -m unittest
Phase-2 Bird handle searches were still spawning Node without the injected AUTH_TOKEN/CT0 env. That left the search pipeline vulnerable to Chrome keychain prompts whenever a query drilled into X handles.
Pass the popup-safe subprocess env through those handle searches and cover it with a regression test.
Classify prompting and animation queries as how_to so the stack does not treat them as generic breaking news. Also keep X available for how_to and preserve YouTube/HN coverage for breaking-news and prediction queries.
Validated with uv run python -m unittest tests.test_query_type and the five-query local comparison run used for PR #65 review.
Update README, launch copy, and UI guidance to prefer popup-free AUTH_TOKEN/CT0 configuration, and keep X backend selection on the verified Bird or xAI paths.
Validation: uv run python -m unittest tests.test_env_project
- Remove duplicate detect_query_type from query.py (divergent 5-type version);
canonical 7-type version lives in query_type.py
- Fix reddit.py import to use query_type.detect_query_type
- Clean unused STOPWORDS/SYNONYMS/tokenize imports from youtube_yt, instagram,
tiktok, scrapecreators_x, bird_x after relevance consolidation
- Fix _relevance_filter default from 0.7 to 0.0 (items without relevance
should not silently pass the filter)
- Remove --dateafter from yt-dlp (returns 0 results for evergreen topics)
- Remove restrictSearchableAttributes from HN search (misses Ask/Show HN)
- Lower HN points filter from >5 to >2 (avoids filtering niche posts)
- Add error logging to select_openai_model HTTP failures
- Remove mise.toml and internal planning doc from repo
- Update module docstrings to describe current purpose, not migration history
- Update tests to import from canonical relevance module
Filter items with relevance < 0.3 per source after dedup, but only
when list has >3 items. Extends the Reddit-only minimum-result
guarantee to all sources: keeps top 3 by relevance if all filtered.
This works with the computed relevance scores from the previous commit
to actually remove off-topic results from the final report.
- hackernews: use extract_core_subject instead of raw topic, add
points>5 filter and restrictSearchableAttributes=title to reduce
noise from URL-match and low-signal posts
- youtube: add --dateafter parameter to yt-dlp for server-side date
filtering (Python soft filter still handles fallback)
- reddit: skip opinion/review query variant for how_to/comparison
queries where it adds noise
- bird_x: add OR-group retry with compound terms before falling back
to word-dropping (uses X OR operator for multi-concept queries)
- query.py: add detect_query_type() and extract_compound_terms()
Matt Van Horn