Add approval-gated SQL writes without exposing DATABASE_URL
September 19, 2026
Allow sql.read, park sql.write for human approval, deny sql.ddl. Agent never sees the connection string.
Read moreBlog
Release notes, engineering deep-dives, and community stories.
September 19, 2026
Allow sql.read, park sql.write for human approval, deny sql.ddl. Agent never sees the connection string.
Read moreSeptember 17, 2026
Production DATABASE_URL in Cursor MCP plus a bad LIKE predicate. @limetry/sql denies sql.write before the database executes.
Read moreSeptember 12, 2026
Run dryRun for a week on live traffic, confirm zero false denies, then move the Admin token behind the firewall.
Read moreSeptember 10, 2026
With @limetry/shopify in path: small refunds allow, larger ones park for approval, duplicates never hit Shopify.
Read moreSeptember 5, 2026
Run evaluate as a non-required job for one week, verify classifications in audit, then make the check required.
Read moreSeptember 3, 2026
Twelve Copilot PRs hit pull_request_target beside configure-cloud-credentials. Evaluate denied deploy before production credentials minted.
Read moreAugust 14, 2026
Use @limetry/sql to classify, evaluate, and execute SQL statements while keeping DATABASE_URL out of the agent.
Read moreAugust 5, 2026
Use @limetry/ci to evaluate ci_privilege and deploy intents before workflows load secrets or production credentials.
Read moreJuly 25, 2026
Hold SHOPIFY_ADMIN_TOKEN in @limetry/shopify, evaluate refund, discount, and inventory intents, and proxy to Shopify only on allow.
Read moreJune 28, 2026
Add a workspace skill that instructs the assistant to run Limetry evaluate before paid-platform side effects.
Read moreJune 11, 2026
Register @limetry/mcp so Claude, Cursor, and Copilot can call evaluate, record, audit, and policy tools.
Read moreJune 2, 2026
Evaluate each sensitive request as an ActionIntent against shared marketplace policy and privacy-safe audit—without holding funds.
Read moreMay 31, 2026
Call evaluate inside the handler before the scale API so channel budget policy can deny oversized cloud upgrades.
Read moreMay 21, 2026
Configure max_cost_minor and allowed action types so purchase-shaped tool calls evaluate before they run.
Read moreMay 10, 2026
POST evaluate from Node or any remote runtime. Point baseUrl at your self-run Limetry server.
Read moreApril 14, 2026
Evaluate agent tool calls against ActionPolicy, return allow/deny with reasons, and store a privacy-safe audit trail. Self-host the open source evaluation server.
Read more