-
SYNAPSE v2.3.7-dev Pre-release
released this
2026-05-11 14:25:38 +02:00 | 128 commits to main since this releasev2.3.7-dev
Added
- Central hardening policy engine (AgentHardeningPolicyService) with unified ALLOW/WARN/BLOCK decisions and reason codes.
- Delegation guardrails for self-target prevention, loop detection, and max-hop enforcement.
- Planning guardrails for max steps and max refinements.
- Token-budget guardrails with concise-mode enforcement when near phase budget limits.
- Policy metadata in native tool execution responses (hardeningReasonCode, nforcedMode).
- New ideas docs:
- ideas/RULES_SYSTEM.md
- ideas/ADMIN_REQUEST_NOTIFICATION_CHANNEL.md
Changed
- Collaboration, planning, and native tool execution now share the same central policy evaluation path.
- Added synapse.hardening.* runtime configuration for delegation/planning/token thresholds.
Notes
- Local Maven test execution remains blocked in this shell (mvn not installed).
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads