Unclear the finding doesn't fully specify which plan-mode scenarios are newly covered, only that the gating was broadened
What
- A new module-level constant (
Cdn, currently hardcoded totrue) is now combined with plan-mode checks to broaden when certain artifact actions — likeverifyandupload_asset— require per-action confirmation rather than being covered by a session-wide approval. This matters because plan mode can admit third-party text into the conversation. - The helper
AA(e)that makes this decision resolves toCdn && ue(e).mode === "plan", and is used together with another helper (Au()) throughout the new artifact consent code, including deciding whether comments or reads need a live confirmation prompt. - A sibling helper,
vU, resolves to(e.ask && e.basis === "unattended") || (EDe && ue(t).mode === "plan"), whereEDeis a second constant currently set totrue, gating a related plan-mode branch.
Why Both gating constants are hardcoded on rather than driven by a real setting, so today this behavior is effectively always active in plan mode: artifact verify/upload/comment/read actions get an explicit confirmation prompt rather than relying on a broader session approval, since plan mode can bring in untrusted third-party text.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
the finding doesn't fully specify which plan-mode scenarios are newly covered, only that the gating was broadened