Community · E2 · artifact verified
Gate agent posts for leaks by audience
A guard for OpenClaw agents that reads an outgoing message and where it is going: Jev judges whether a client name, credential or internal hostname is about to reach the wrong readers, then confirms, blocks or rewrites per channel.
01 · Role in the system
What Jev does here
Before an agent's status update lands in a channel, the guard sends Jev both the text and the destination. The same update naming a client and where its credentials live gets a confirmation for the private ops channel and a block for the channel shared with a vendor - the judgment is message-times-audience, not a keyword blacklist. Decisions come from the systemone wire protocol against a configurable backend (hosted Jev or a local decision model), each call measured with its latency, and the README ships bilingual docs, a demo gif of the same post allowed and blocked, and compatibility pinning for OpenClaw 2026.9.6+. It is the outbound complement to incoming-message moderation: the question is not whether text is toxic but whether this reader should see it.
02 · Control boundary
Where Jev sits
Jev as the outbound leak gate on agent messaging: message and destination are judged together over the systemone wire protocol, per-channel policies turn typed answers into confirm, block or rewrite, and latency is measured per call.
Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.
03 · Known limits
What this evidence does not prove
- Built for OpenClaw 2026.9.6+; other agent frameworks need their own hook.
- Leak categories (client names, credentials, hostnames) are the configured question set; novel leak shapes need new questions.
- Brand-new repo (created 2026-09-30, 0 stars); no independent evaluation.
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- yousan ↗Community · github · public · checked 2026-09-30