ENGINEERING

Block high-risk dependency bumps until a senior reviewer signs off

Grades each dependency-bump PR from its changelog and, when the risk is high, posts a blocking review and assigns a senior reviewer so the bump cannot merge unattended.

CategoryEngineering
Enginesim
Difficultyintermediate
Triggerevent
Steps6
Setup~15 min

How it runs

The automated pipeline, trigger to output.

  • TriggerDependency-bump PR opened or updatedGitHubGitHub
  • ActionFetch changelog for the version spanGitHubGitHub
  • ActionLLM grades risk + lists breaking signalsOpenAI
  • LogicContinue only if grade is high
  • ActionPost blocking REQUEST_CHANGES reviewGitHubGitHub
  • OutputAssign senior reviewer from rotationGitHubGitHub

What it does

This workflow turns the risk grade into a merge gate. It reads the changelog for the version jump, classifies the breaking-change risk, and only intervenes on `high` results — posting a `REQUEST_CHANGES` review, assigning a senior reviewer, and explaining why in the review body. Low and medium bumps pass through untouched.

When to use it

Use it when you trust most automated bumps to auto-merge but want a hard stop on the dangerous ones — major version jumps, removed APIs, or framework migrations — so they never slip through an auto-merge rule.

How it works

  1. 1A PR-opened or PR-synchronized webhook fires on dependency-bump PRs.
  2. 2The flow extracts the package and the old→new version span and fetches the changelog.
  3. 3An LLM grades the risk and lists the specific breaking signals it found.
  4. 4A branch checks the grade: anything below `high` ends the run quietly.
  5. 5For `high` PRs, a blocking GitHub review is posted with the rationale.
  6. 6A senior reviewer from a configured rotation is assigned as the final output.

Set it up

What you configure once, before turning it on.

  1. 1
    Connect GitHubRepos, issues, pull requests, actions.
  2. 2
    Connect OpenAIModels, embeddings, files.
  3. 3
    Set each agent's modelWe leave models unset so you pick the tier — fast + cheap, or top-quality.
  4. 4
    Tune it to your dataEdit the prompts, filters, and field mappings so it matches how your team works.
  5. 5
    Test, then turn it onRun once against a sample, confirm the output, then enable the trigger.

Run this workflow in your colony.

14-day trial. No DevOps. No Sales call. Provisioned in under a minute.