Fleet report — who ran, what they brought
A live view of the multi-agent fleet: each agent's identity (tool), status, runtime, and what it contributed. This page polls every 5s — run python agents/run_scout.py all and watch it update.
Intelligence pipeline
→ competitors.json · trends.jsonHigh-recall gather → adversarial verify → high-precision synthesis. Runs per target (each competitor / trend).
Searches the web from many angles and extracts concrete, checkable claims.
Runs INDEPENDENT corroboration searches per claim and grades it (the governance step).
Synthesizes only verified signal into decision-grade records, through a JFrog lens.
Signal collectors
→ signals.jsonThe antenna — parallel collectors pull community & vendor chatter, then one analyst clusters it into themes.
Hacker News stories by topic.
Posts from dev/security subreddits.
X threads, blogs, forums.
Vendor & security blog feeds (JFrog, Snyk, GitHub, BleepingComputer…).
Clusters the raw chatter into themes with a JFrog implication each.
Threat feed
→ threats.jsonDeterministic — no LLM, no flaky scraping, so the live feed never goes dark.
Recent reviewed advisories per package ecosystem.
Vulnerabilities known to be actively exploited.
Media generation
→ /media (mind map · videos · podcast)Turns the same research sources into explainer media.
Generates a mind map, per-product/competitor video overviews, and a podcast.
Last run — agent by agent
No run recorded yet. Run the pipeline to populate.
Yield — measured, not vibes
The Verifier's confidence ladder
≥2 independent sources clearly support it
only one source supports it
evidence thin or tangential
evidence contradicts it — surfaced, not hidden