Release notes
1.0.2#
Changed
- The bundled Leet Guardrails documentation now states the current schema version (2); the code was already on it.
- Development package metadata is now published under Technoleet (was Ankabut Software). Nothing shipped to sites changes.
1.0.0 (2026-11-01)#
Initial release.
New
- Server-side gate for ability calls from agents: Application Passwords, AI-plugin agent users, MCP STDIO sessions and in-process AI Client calls.
- Deterministic risk scoring for every ability, with an optional Ability Auditor signal, on the Risk screen.
- Allow, hold for approval or deny, by risk default, agent mode and rules.
- Approval queue under Agent Control > Approvals; approve runs the action as the requesting user; the agent collects the result once by repeating the call or via
ability-gate/get-request. - Request screen with input, risk reasons, row-level changes, external effects and audit events.
- Admin-only Preview changes for held requests, with HTTP, email, cron and schema changes blocked.
- Exact SQL row-level snapshots and one-click rollback with drift check; external effects listed and partial snapshots marked.
- Agent modes (paused, restricted, trusted), rules for abilities, namespaces and categories, time-boxed grants and hiding of denied abilities.
- Per-agent hourly limits on calls and writes.
- Hash-chained audit trail with retention, CSV and NDJSON export and integrity check.
- Approver emails with a 10-minute digest and an admin-bar pending count.
- Monthly AI token budgets, site-wide and per plugin, through the WordPress AI Client.
- WP-CLI commands:
wp ability-gate requests,auditandagents. - Leet Guardrails developer API and hooks.
Was this article helpful?
Tell us what was missing
