[v2] Deprecate v1 ReasoningAuditor to Advisory-Only Layer

[SHAURYASANYAL3]

Phase 5: Integration & Migration

The Context

The v1 ReasoningAuditor (which uses an LLM to evaluate agent reasoning) is fundamentally unreliable as a hard security gate. We are removing its blocking authority, but keeping it available for users who want optional semantic advice.

Tasks

• Refactor ReasoningAuditor so that it no longer returns ExecutionStatus.BLOCKED.
• Relabel its output to ADVISORY or WARNING.
• Rip out the legacy regex-based pattern matching in SafetyEngine entirely, as these checks are now formally handled by the Capability and State lattices.
• Ensure tests reflecting the old blocking behavior are updated to expect Lattice-based blocks instead.

Part of Epic #651

[akhilayadavborra]

Hi! @SHAURYASANYAL3 I'd like to work on this issue. I have reviewed the requirements and would like to take it up. Could you please assign it to me? Thank you!

[rohitkumarnaidu]

Hey maintainer, I'd love to work on this issue and submit a PR. Could you please assign it to me?

[rohitkumarnaidu]

Hi! Could you please add the ELUSOC label to this so it gets tracked properly by the dashboard? Thanks!

[arcgod-design]

Hi @SHAURYASANYAL3! I'd love to work on this Phase 5 migration task under SSoC '26. I'm @arcgod-design.

My Approach:

• Refactor ReasoningAuditor to remove ExecutionStatus.BLOCKED returns, replacing them with ADVISORY / WARNING outputs only
• Strip out the legacy regex-based pattern matching from SafetyEngine since those checks are now owned by the Capability and State lattices
• Update all unit/integration tests that previously expected blocking behavior to instead expect Lattice-based blocks
• Verify no v1 blocking paths remain active in the final pipeline

Could you please assign this to me (@arcgod-design)? Ready to start immediately!

[sreerevanth]

@arcgod-design take up put on a clean pr