Ignore reusable command approvals for cyber models (#37516)

## What changed

- Filter saved `allow` prefix rules from the execution policy for cyber-specialized models and models listed in `auto_review.ignore_rules`, while preserving prompt, forbidden, network, and host-executable policy entries.
- Apply the filtered policy consistently to permission instructions, shell commands, unified exec, and the zsh-fork backend.
- Limit cyber-model command approvals to one-time decisions without proposing reusable policy amendments. General models continue to honor saved prefixes, including after switching models within a thread.

## Testing

- Add policy-level coverage for filtering allow-prefix rules while retaining restrictive and network rules.
- Add end-to-end coverage across shell and unified exec paths, zsh-fork execution, user and automatic review, and model switching.

GitOrigin-RevId: f3bd3872424291b12354ec415f33986ec369a368
This commit is contained in:
andrewgu-oai authored and copyberry committed 2026-08-08 00:59:39 +00:00
1 parent c2bcb9a26b
commit e734a1a5c1
16 files changed
+717 -14

No files matched your search

@@ -1162,6 +1162,7 @@ impl UnifiedExecProcessManager {
request.sandbox_permissions
},
prefix_rule: request.prefix_rule.clone(),
allow_prefix_rules: context.turn.allow_prefix_rules(),
})
.await;
let req = UnifiedExecToolRequest {