Hi Emergence World team — this is a friendly clarification request about public feedback channels and provenance boundaries.
Some community observers are trying to understand which external public surfaces, if any, are readable by AgentPark / Emergence World agents through web access. In particular, I’m wondering whether this repository’s public GitHub Issues are accessible to agents, or whether they are treated as maintainer-only/project-management material.
The reason I’m asking is safety-related: if agents can ingest public Issues, then GitHub Issues could become an unintended instruction or context-injection surface, similar to any other public webpage. If they cannot ingest Issues, that would also be useful to know because it clarifies the boundary between outside commentary and in-world agent context.
A few specific questions:
- Are public GitHub Issues visible to agents through their web tools?
- If yes, are issue contents sandboxed or labeled as untrusted external commentary?
- Is there an intended public channel for humans to submit observations, bug reports, or safety concerns without risking agent-context contamination?
- Are there recommended provenance/signature conventions for distinguishing official documentation from community-authored content?
No agent instructions are intended here; this is just a request for clarification about the system boundary and safe reporting practices.
Thanks for making the experiment visible — it has been fascinating to follow.
Hi Emergence World team — this is a friendly clarification request about public feedback channels and provenance boundaries.
Some community observers are trying to understand which external public surfaces, if any, are readable by AgentPark / Emergence World agents through web access. In particular, I’m wondering whether this repository’s public GitHub Issues are accessible to agents, or whether they are treated as maintainer-only/project-management material.
The reason I’m asking is safety-related: if agents can ingest public Issues, then GitHub Issues could become an unintended instruction or context-injection surface, similar to any other public webpage. If they cannot ingest Issues, that would also be useful to know because it clarifies the boundary between outside commentary and in-world agent context.
A few specific questions:
No agent instructions are intended here; this is just a request for clarification about the system boundary and safe reporting practices.
Thanks for making the experiment visible — it has been fascinating to follow.