What it does
CI/CD Pipeline Optimizer goes after the twenty-minute CI wait that quietly costs more than it looks like it does, because the real price is not the twenty minutes but the context switch on either side of it. You supply a workflow's step-by-step timings — the target, the workflow file, the total duration and the per-step durations — and it scores each step against its share of total wall-clock time, then proposes changes from four specific playbooks rather than generic advice. Add a dependency cache where a step has no cache hits and takes over thirty seconds. Decouple a step from the critical path when nothing depends on it. Shard a test step across a matrix when the suite is large enough to be worth splitting. Delete a step whose coverage another step already provides. Every recommendation carries an estimated saving in seconds, and the report leads with the before and after totals — eighteen minutes down to seven, sixty percent faster — computed arithmetically from those savings rather than asserted. The language model narrates the numbers; it never produces them. The deliverable is a workflow diff: the exact YAML edits in unified-diff form, ready to turn into a pull request. Not a list of suggestions to translate into changes yourself. One risk is called out in every report rather than left for you to discover: aggressive caching can produce a false-positive green build. A wrong cache key means the pipeline passes without testing what you changed. Any cache recommendation this agent makes needs its key verified before the PR merges, and it says so every time rather than presenting the time saving alone. Each step is callable on its own. `breakdown_steps` answers where the time goes. `estimate_savings` gives you the before-and-after proof for a status update. `preview_workflow_diff` returns just the YAML edits. `run_full` does everything. No connectors and no credentials — it runs on the timings you provide and opens no pull requests itself. Built for DevOps engineers and tech leads who own the pipeline everyone waits on.
Example prompts
- Here are our CI step timings — how do we halve the runtime?
- Which steps are the bottleneck in this pipeline?
- Where should we add caching and how much would it save?
