A Claude Code plugin that turns a plain-English request into a Blink automation. It drafts the workflow YAML, validates it against your workspace's real action catalog, saves it as a draft in Blink, and with your approval test-runs and publishes it.
"Every morning at 8am, check for EC2 instances without an owner tag and post the list to #cloud-ops."
/plugin marketplace add blinkops/blink-code-builder
/plugin install blink-code-builder@blink-marketplace
Then ask Claude to set up the plugin. The bundled setup-blink-plugin skill handles the rest:
Python prerequisites, walking you through the four config values (/plugin → Configure options),
and verifying the catalog populated.
After setting the config values, quit Claude Code and reopen it — they don't take effect mid-session.
You'll need a Blink user API key, and Contributor or above in the target workspace. The company-level "Builder" role alone carries no workflow permissions.
The bundled blink MCP server reads your connections and playbooks, saves drafts, triggers test
runs, fetches run logs, and publishes.
A test run is not a dry run. It sends the real messages, opens the real tickets, and deletes the real records. The plugin blocks a run it considers risky and asks you first, but the checks are heuristics, so your review is what actually keeps a run safe.
What you need to do:
- Keep
trigger_test_runandpublish_automationout of your permission allowlist. Claude Code's prompt is the real enforcement. The plugin's own gates are unblocked by a flag Claude passes, and nothing verifies a human said yes. - Maintain
automations/connections-allowlist.yaml. Only connections you list there can be used in a test run. If the file is missing, no run is allowed at all. - Read what Claude tells you before approving. When a run is blocked, Claude explains who and what the flagged steps touch. That summary is the decision point.
- Think twice before approving a publish. Publishing needs your approval, and once a workflow is published its schedule and its triggers are live. A scheduled or event-based automation will run on its own, without anyone watching it.