Agents · Support Agent
Support Agent
Slug support. Confidence threshold 0.70. No policy pack attached.
Quality & regression
Wires the M6 B2 regression-on-save gate. When a non-off policy is paired with an active golden set, every save to this agent (and to its KB, tools, or policy pack) queues aconfig_save eval run. The save commits or reverts based on the run's result.
Require approval on save
Policy pack
Pick a pack to apply topic rules + refusal templates + optional confidence-threshold override on this agent's turns. Detach by selecting "(no policy pack)".
Persona
first-line policy support
Allowed risk classes
Sets which tool risk classes this agent is permitted to invoke. Tools whose risk class is not allowed are filtered out of the LLM-facing menu AND refused by the executor (belt and braces). Default is info-only — widen explicitly to grant change or sensitive access.
Attached tools
The LLM only sees tools attached to this agent. Save replaces the attachment set atomically. A risk class badge per tool flags when a sensitive tool is attached to an agent that is not permitted to call it — useful early-warning when provisioning.
Versions
Every save snapshots the agent. Restoring an older version appends a new entry — history is append-only.
| # | Note | When | |
|---|---|---|---|
| v3current | restored from v1 | 5/12/2026, 3:59:24 PM | |
| v2 | tools attached | 5/12/2026, 3:59:21 PM | |
| v1 | tools attached | 5/12/2026, 3:59:17 PM |