Code
T9105 policy-refused
What a project tool was asked for, or what a workspace holds, is something that the product does not let through: the policy of the agent does not allow it, a path leads out of the workspace, or code may not run there.
Details
| The case | What is refused |
|---|---|
| A path | A path that leads out of the workspace's own project, also through a link or by other capitals, and a path of .tent/, tent.clash.ts, tent.d.ts and mcp.lock.json, which the framework writes itself, and engine.ts where also does not name 'engine' |
| A write that the policy refuses | A file that a tool wrote and that holds what the policy of the agent does not allow: a definition of a kind that create, edit or remove does not name, as a route, or a seed.ts of code where no kind names seed, an option that only also opens, or a host outside reaches. The file is as it was before the tool |
| Code that nothing lets run | A workspace, which is code, where neither a role of its owner that trusted names nor a sandbox lets it be built and run, and the publish of such a workspace into the product's own engine: it is published to a far server alone |
A model that a tool answered T9105 is handed the rule that was broken in fields, under rule, so it can write the change again within the policy.
The fix
Ask for what the policy allows, within the workspace. A product that wants its AI to do more widens the policy of its agent; one that runs code of its users connects a Sandbox, or names the roles it trusts in trusted.