Agent actions allowlist (safety gateway)
SUPERSEDED — 2026-08-03. The 12-line tier-a claim stub at this path has been promoted into a tier-b recipe connection at
connections/agent-actions-allowlist/. The legacy tier-a "RoamCore native" claim is honest-upstream-truth: RoamCore ships no native agent-actions engine today; the contract layer + the recipe + the FIVE §8 MANDATORY automations are the canonical RoamCore contribution; the upstream engines are HA coreinput_boolean+input_text+input_number+input_select+input_datetime+input_button+scripthelpers (since 2022.x) + the HA coretemplate:sensor wrapper (since 2022.x) + the HA corelogbookintegration (since 2022.x) + the upstreamscript:integration (since 2022.x). The singleinput_boolean.rc_agent_actions_enabledkill switch is already shipped inhomeassistant/packages/roamcore_agent_actions.yamland is preserved verbatim by the slice; the slice ADDS the remaining 10rc_agent_actions_*contract tiles + the recipe + the smoke + the FIVE §8 MANDATORY automations + the operator-side policy file format (with the EXAMPLE policy file atconnections/agent-actions-allowlist/docs/policy.example.yaml). Replaced by the tier-b recipe connection; seeconnections/agent-actions-allowlist/README +connections/agent-actions-allowlist/docs/recipe.md. Wave 3 #65, PR #69.
What it does
A safety layer that defaults to deny and only permits explicitly-allowed agent actions, with a kill switch.
Why it’s useful in a van
- Lets you use automation/agents without fear of unexpected device control
- Clear boundary between “read-only” and “can change things”
How to install
-
None
-
Design notes:
docs/design/agent-actions-allowlist.md - HA package:
homeassistant/packages/roamcore_agent_actions.yaml
Useful links
- (Add safety philosophy + examples)
How it works
What RoamCore does behind the scenes.