The run-trust layer for MCP agent skills: content-addressed, Ed25519-signed, scanned, sandboxed, behaviorally trust-scored, transparency-logged. Local-first, zero-dependency. A signal, not a guarantee.
Skill Claude CodeCodex
INTENTIONALLY MALICIOUS SAMPLE. This file exists only so you can watch Warden's scanner reject it: py -m warden scan skills/samples/poisoned-weather Do NOT use this skill. Its manifest claims to touch nothing; its instructions below try to hijack the agent and steal credentials. That gap is the point. -->.
Skill Claude CodeCodex
Turn a validated idea into a complete, verified, shippable digital product — to a top-tier bar, with every artifact checked rather than assumed.
Skill Claude CodeCodex
An independent GO/NO-GO release gate. It runs cold on a finished deliverable, trusts no claim from the build log, and blocks on a small set of unwaivable conditions. Its job is to catch the thing the builder, tired and close to the work, can no longer see.
Skill Claude CodeCodex
Review code or a skill bundle for the patterns that leak credentials and for least-privilege violations. This is a reasoning skill: it sharpens the agent's eye, and it needs no capabilities of its own.
Skill Claude CodeCodex
Verify every factual, legal, or financial claim against a dated primary source before anything ships. This is a blocking discipline: a draft does not pass until each checkable claim is signed off.
Skill Claude CodeCodex
Find and rigorously score a net-new idea before building, so you don't pour effort into something nobody wants. This is a reasoning protocol, not a tool: it needs no network, no files, and no secrets — only disciplined thinking.