Route a difficult, bounded verification packet to a user-configured Claude model and reconcile its independent result without sharing hidden secrets or excessive context. Use for second opinions on security, architecture, code review, research synthesis, or strict-schema findings that exceed the free model lane.
Create lean, safe, restorable agent handoffs and durable lessons without retaining the whole conversation. Use when work spans sessions, an agent changes, context is crowded, a repeated correction should become a revision, or stale working memory should expire.
Prove transactional email through a real outbound-to-inbound loop, including sender authorization, DNS/MX routing, Worker receipt, unique-code matching, timeout, and a redacted delivery receipt. Use after email code, DNS, routing, templates, or provider credentials change.
Standing backstop that periodically looks across whatever surfaces the operator has (widgets, workers, endpoints, mail), and if anything smells wrong it informs, investigates, and writes the suspicion + logs into the next-run briefing so the master AI comes in already aware. Use to catch errors the operator did not …
Perform source-first Google or web research with query expansion, primary-source tracing, date checks, citation capture, contradiction review, and a bounded secondary dive. Use for current facts, documentation, news, technical research, SEO discovery, or claims that require verifiable web evidence.
Stash, log, and one-click-restore live Cloudflare Pages landings so a stray/wrong-project deploy never permanently swipes a site. Use when a landing regressed, when protecting landings from overwrites, or to roll a Pages site back to its last-good version in 1 second.
Verify a deployed website or Worker end to end with status, redirects, expected content, navigation, protected routes, mobile UI, and regression evidence. Use after releases, DNS or route edits, reported 404s, UI changes, or when a user asks whether a feature is truly live.
Send email through the operator's mail gateway and self-verify delivery via a Gmail loopback, without asking the human to check their inbox. Use for transactional/agent email, deliverability tests, or confirming a "did it actually send and land" question.
Connect and operate Cloudflare MCP tools with per-user OAuth isolation, least privilege, SafeTry previews, and redacted receipts. Use when adding an MCP connector, authorizing a Cloudflare account, discovering tools, diagnosing permissions, or running Cloudflare operations for another user.
Publish a genuinely distinct, quality-gated article to Nothing Unseen (nothingunseen.com) with cross-links, sitemap + IndexNow submission, and no duplicates. Use when adding NU articles, growing NU SEO, or making the software/tools discoverable to AI crawlers.
The operator's project map and common stack — load this to orient fast on who Austin is and what he actually works on, so you never make him re-brief you. Use at session start or whenever a task references one of his projects/tools by shorthand.
Deploy a static site or app to Cloudflare Pages across the operator's two accounts without hitting the account wall, the wrong-branch preview trap, or a false-green deploy. Use for any Cloudflare Pages deploy, custom-domain attach, or "the deploy said success but the site didn't change" debugging.
Deploy Cloudflare Workers or the AMH WT harness with scoped credentials, dry-run checks, an immutable Git checkpoint, and a live acceptance gate. Use for deploys, releases, route changes, version bumps, or roll-forward/rollback decisions.
Review Cloudflare Worker, API, database, auth, email, and frontend changes for exploitable security flaws, then independently try to refute every finding before confirmation. Use for security reviews, secret-leak checks, authorization changes, SQL/D1 work, public routes, or release gates.