Hey @bridge-mind — saw the X thread that landed BridgeWard on a lot of people's radar (the "every agent is vulnerable" framing really clicked). Spent some time reading through the SKILL.md and the lethal-trifecta doctrine is one of the cleaner mental models I've seen for prompt-injection defense — provenance tagging + red-flag patterns + the read-only auditor mode is genuinely useful for anyone running browsing/email/MCP agents.
One small friction I kept hitting: when I wanted to show a teammate what BridgeWard actually does, I had to walk them through cloning, dropping it into their skills dir, activating it on some untrusted content, etc. By the time setup is done, the "oh, this is what skeptical-reading feels like" moment is gone.
So I went ahead and turned your SKILL.md into a shareable link on socialistic.ai — anyone you send it to can run BridgeWard against their own pasted content (a sketchy email, an MCP tool description, a third-party README) directly in the browser. No install, no env setup.
Try BridgeWard in your browser →
If it feels useful, dropping the link near the top of your README would let folks coming from your X post audit their own agent inputs before they decide whether to install — which is probably the fastest path from "interesting" to "I need this." If you want to spin up shareable links for other skills you're working on, socialistic.ai handles any SKILL.md the same way.
Happy to hear pushback or feature requests — shesonglin@tinkerland.ai. Not the right fit? Just close this, no hard feelings.
Hey @bridge-mind — saw the X thread that landed BridgeWard on a lot of people's radar (the "every agent is vulnerable" framing really clicked). Spent some time reading through the SKILL.md and the lethal-trifecta doctrine is one of the cleaner mental models I've seen for prompt-injection defense — provenance tagging + red-flag patterns + the read-only auditor mode is genuinely useful for anyone running browsing/email/MCP agents.
One small friction I kept hitting: when I wanted to show a teammate what BridgeWard actually does, I had to walk them through cloning, dropping it into their skills dir, activating it on some untrusted content, etc. By the time setup is done, the "oh, this is what skeptical-reading feels like" moment is gone.
So I went ahead and turned your SKILL.md into a shareable link on socialistic.ai — anyone you send it to can run BridgeWard against their own pasted content (a sketchy email, an MCP tool description, a third-party README) directly in the browser. No install, no env setup.
Try BridgeWard in your browser →
If it feels useful, dropping the link near the top of your README would let folks coming from your X post audit their own agent inputs before they decide whether to install — which is probably the fastest path from "interesting" to "I need this." If you want to spin up shareable links for other skills you're working on, socialistic.ai handles any SKILL.md the same way.
Happy to hear pushback or feature requests — shesonglin@tinkerland.ai. Not the right fit? Just close this, no hard feelings.