Lock the last 7 days as the default time window and cover official blogs, developer communities, and X discussion signals within the same run.
⤴
PARALLEL EVIDENCE TRACKS
Decision Node
Make the default workflow explicit as two parallel evidence tracks rather than a single linear path: one for public sources and one for X signals.
⚙
EXPAND PUBLIC-SOURCE EXPLORATION
Processing Node
In one evidence track, explore Hacker News, GitHub, RSS feeds, official blogs, web search, and official documentation.
⚙
EXPAND X EXPLORATION
Processing Node
In the second evidence track, expand X sampling: reuse and validate the `/tmp` cookie first, then follow the timeline-first collection order.
⚙
COLLECT STABLE PUBLIC SOURCES
Processing Node
Prioritize stable sources that do not require extra accounts, and treat official announcements, clearly dated product updates, and highly discussed implementation notes as high-signal samples.
⚙
RUN X IN THIS ORDER
Processing Node
Run the X track in the documented order: reuse and validate the cookie, sample high-signal timelines, add Home feed only when insufficient, and use narrow search only as the final fallback.
⚙
PRIORITIZE HIGH-SIGNAL PUBLIC INFO
Config Node
Prioritize official announcements, product updates with clear dates, highly discussed community posts, and implementation notes or developer writeups.
⚙
APPLY X COOKIE RULES
Processing Node
Use only `/tmp/last_7_days_news_x_cookie.txt`, verify that `auth_token` and `ct0` are present, and do not trust the file just because it exists.
⤴
REFRESH LOGIN IF NEEDED
Decision Node
If the current cookie is invalid, run `scripts/ensure_x_cookies.sh`, complete login in the opened browser, then export and validate the refreshed cookie again.
⚙
SAMPLE HIGH-SIGNAL TIMELINES
Processing Node
Sample model companies and official accounts first, then researchers, then the AI coding and agent-tool ecosystem; do not start from the search page.
⤴
MERGE, CROSS-CHECK, AND RANK
Decision Node
After both the public-source track and the X track finish exploration, merge the results, deduplicate them, cross-check claims, and rank by source quality, freshness, engagement, and relevance.
⚙
BUILD MERGED CANDIDATE SET
Processing Node
Combine the normalized public-source results and X findings into one merged candidate set before the final summary layer.
⚙
USE HOME / SEARCH ONLY AS FALLBACK
Processing Node
Use Home feed only when timeline samples are insufficient; try narrow search only when both timeline and Home feed are insufficient, and return to timelines immediately if search fails.
⟲
PRESENT HOT TOPICS / VIEWPOINTS / OPPORTUNITIES
Output Node
Present `Hot Topics / Viewpoints / Opportunities` on top of the merged result layer, and generate a one-page summary or HTML report when a shareable artifact is needed.