Tools

Agent org control plane. Verified working on the overnight bench — boots clean, embedded Postgres, telemetry off.

Telegram-gateway agent runtime. My daily driver.

Persistent agent harness. Installed on a Factory droid backend; first real run still pending.

Won my head-to-head demo comparison against two alternatives; full re-scaffold pending. Promising.

Lost the same head-to-head to Cabinet on my use case.

Desktop app builds and boots clean end-to-end (Bun + Electron). Works; not yet part of the daily stack.

Block's agent workspace. Cloned, never judged — the source build wants multi-GB toolchain and disk ran out mid-bench. Prebuilt release is the easy path.

Weaker take on what Buzz does, with a restrictive license on top.

Plan-cheap-execute optimization loop as a Claude skill. Cherry-picked from the bench.

Three skills kept after a security read: risky-changes (adapted), folder-specific-claude-md, effective-agent-skills.

Design-research skill pack with a curl|bash installer; paid tier can decrypt browser cookies. Only ever deliberately, with the cookie-decryption caveat in mind.

Browser automation CLI for agents — accessibility-tree snapshots at a fraction of screenshot token cost. Daily driver.

Scrape/crawl/extract API for pages that fight plain fetch. Its X postprocessor is the only reliable way I read tweets programmatically.

Score how ready a website is for AI agents, then get evidence and recommendations to improve it.

Token-saving MITM proxy with documented silent string corruption. Unusable anywhere fidelity matters.

Unsigned app from a personal tap — not worth the trust exception.

Duplicates my existing markdown vault workflow.

The workspace product turned out to be a Notion replacement, not the PDF tool I wanted.

Fine tool, but requires a Rust toolchain I don't want to carry right now.

Vercel's agent sandbox. Scaffolded and ready to test — needs Node 24.