feat(blog): add 'Anyone in the world can put text in front of your agent's model for the price of an email'

[jiashuoz]

What

Adds today's blog post, "Anyone in the world can put text in front of your agent's model for the price of an email", following the existing blog pattern:

• web/src/app/blog/inbound-prompt-injection/page.mdx - the post (~520 words)
• web/src/app/blog/posts.ts - registry entry (date 2026-09-06, 3 min read); sitemap picks it up automatically

Angle

An agent that reads its inbox and acts on it is reading untrusted input the whole internet can write - inbound email is a prime indirect prompt-injection vector (hidden HTML, zero-width / Unicode-tag smuggling, encoded payloads). The "system prompt says ignore it" fix is a hope, not a control. The post walks e2a's opt-in inbound content screening: dependency-free heuristics detector (OWASP LLM01 / MITRE ATLAS mapped) plus optional LLM detector adding semantic injection and phishing classification, allow / review / block verdicts by scan sensitivity, review feeding the same HITL queue as outbound holds, fail-safe to review (never silent allow), and verdicts audited in protection_events. Names the beta status of the protection surface openly. Every claim grounded in the repo README; no competitor named; no efficacy claims.

Verification

• npm run build (static export) succeeds; the new page prerenders at /blog/inbound-prompt-injection
• Visually verified by screenshot: the post page renders in the site's existing style and the post appears at the top of /blog

Not merged - review and merge when ready. Note: prod serves a pinned release, so the post won't be live until a release is cut and promoted.