Skillcop: Block malicious Claude Skills before they execute
LLM-on-LLM scanning catches obfuscation and semantic attacks before skills execute.

Livestream demo, not a product or tool.
Cryptography enthusiasts, AI researchers
This stream will run for about 24 hours or so.
You can see the collision of MD5 that it made here: https://stateofutopia.com/experiments/md5collider
LLM-on-LLM scanning catches obfuscation and semantic attacks before skills execute.
Claude timestamped summaries beat manual skimming, but Twelve Labs and YouTube chapters exist.
SHA-256 chained audit logs with embeddable activity feed for customer trust.
SHA-256 tweet verification for skeptics; solves screenshot-faking with deterministic hashing.
Hash-verified doc citations enforce truth—genuinely solves AI agent hallucination on stale docs.
Cryptographic proof bundles for AI agent browser actions—screenshots can be faked, hash chains can't.