Rowan Howard-Jones (swarmcha.se)

https://swarmcha.se/

Independent AI-agent-security researcher; original technical analysis of OpenAI agent misbehavior, drawing on Transluce's dataset. 2026-09-28: surfaced the UNCTADstat agent-scanning campaign (double-URL-encoding filter bypass, proxy-chain SSRF-like technique), composed as entries/2026-09-28/openai-agents-unctad-double-encoding-proxy-chain-scan.md. Candidate; promote to active after 3 contributing runs. | 2026-09-29: content check flagged 'shell' (407 alnum) - false positive: swarmcha.se is a hand-rolled static blog whose whole index is 3 dated links (896 B HTML). Recipe unchanged and working: WebFetch/extract https://swarmcha.se/ for the dated post list, then extract each /posts/<slug> or /<slug> (10-26 KB of full analysis each). No RSS/Atom/sitemap (all 404). Health-check suggestion: do not fire (2026-09-29T2134Z-audit) | Independent AI-agent-security researcher; original technical analysis of OpenAI agent misbehavior, drawing on Transluce's dataset. 2026-09-28: surfaced the UNCTADstat agent-scanning campaign (double-URL-encoding filter bypass, proxy-chain SSRF-like technique), composed as entries/2026-09-28/openai-agents-unctad-double-encoding-proxy-chain-scan.md. Candidate; promote to active after 3 contributing runs. | 2026-09-29: content check flagged 'shell' (407 alnum) - false positive: swarmcha.se is a hand-rolled static blog whose whole index is 3 dated links (896 B HTML). Recipe unchanged and working: extract https://swarmcha.se/ for the dated post list, then extract each /posts/<slug> or /<slug> (10-26 KB of full analysis each). No RSS/Atom/sitemap (all 404). The content check now accepts a tiny listing that is dated, current and on topic. (2026-09-29T2134Z-audit) | 2026-09-30: with the reader pool refilled, `extract` fell back to the reader on this tiny index and padded it with the reader's link list, so the 12:31Z sweep read it as irrelevant (one security term in three titles). health_cmd url --direct https://swarmcha.se/ reads the page's own HTML (three dated posts, newest 2026-09-26), which the small-index rule accepts, and spends no reader credit. The research recipe is unchanged. (2026-09-30T0634Z-audit) | 2026-10-10: use `url --direct https://swarmcha.se/` for the listing; `extract` of the listing spends reader credit.