What the agent does
The agent reads the chart and its neighbours before it adds a template. In a repository like prometheus-community/helm-charts, the prometheus-redis-exporter chart at charts/prometheus-redis-exporter is at version: 6.35.0 with appVersion: v1.95.0, and its templates/ directory holds deployment.yaml, service.yaml, serviceaccount.yaml, servicemonitor.yaml, prometheusrule.yaml, networkpolicy.yaml, role.yaml, rolebinding.yaml, extra-manifests.yaml, NOTES.txt and _helpers.tpl, with no disruption budget. values.yaml sets replicaCount: 1, and deployment.yaml takes replicas from it and its pod selector from the prometheus-redis-exporter.selectorLabels helper. The chart's README.md documents its values, and ci/extramanifests-values.yaml is the one extra values file chart-testing installs with.
The same repository already has the pattern it wants. charts/prometheus-blackbox-exporter/templates/poddisruptionbudget.yaml is wrapped in {{- if .Values.podDisruptionBudget -}}, emits policy/v1 PodDisruptionBudget named with the chart's fullname helper, labelled with its labels helper, selects on its selectorLabels helper, and copies whatever the podDisruptionBudget value holds into spec with toYaml, so a user sets maxUnavailable or minAvailable and nothing else is invented. Its values.yaml has podDisruptionBudget: {} with a commented maxUnavailable: 0. The repository's CONTRIBUTING.md requires a semver version bump for any chart change, and .github/workflows/ci.yaml runs ct lint with .github/linters/ct.yaml, fails when more than one chart changes in a pull request, and installs each changed chart on a cluster once per values file under its ci/ directory, with helm unittest --strict as an additional command for charts that have a unittests/ directory.
The agent copies that pattern into charts/prometheus-redis-exporter/templates/poddisruptionbudget.yaml, using this chart's own fullname, labels and selectorLabels helpers from _helpers.tpl, adds podDisruptionBudget: {} to values.yaml with the same commented example, and documents the value in the chart's README.md table. It bumps Chart.yaml from 6.35.0 to 6.36.0, a minor bump for a new optional feature, and adds ci/pdb-values.yaml setting replicaCount: 2 and podDisruptionBudget.maxUnavailable: 1, so chart-testing installs and tests the new template on every pull request from now on. It adds unittests/poddisruptionbudget_test.yaml with two cases, one asserting no document renders by default and one asserting the maxUnavailable and selector when the value is set, because ct.yaml runs the plugin on any chart that has the directory. It touches no other chart, since ci.yaml would fail the pull request otherwise.
Before pushing, the agent runs helm lint on the chart, helm template with the default values to show no budget renders, helm template -f ci/pdb-values.yaml to show the rendered PodDisruptionBudget with the right selector, helm unittest --strict with the new test file, and ct lint --config .github/linters/ct.yaml against the branch, which also validates the version bump and the README. It opens the pull request with those outputs and a note that a budget with replicaCount: 1 and maxUnavailable: 0 would block node drains, which is why the default stays empty.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Add an opt-in PodDisruptionBudget to the prometheus-redis-exporter chart in this Helm charts repository and open a pull request. Read charts/prometheus-redis-exporter/Chart.yaml, values.yaml, README.md, templates/deployment.yaml, templates/_helpers.tpl and ci/, then read charts/prometheus-blackbox-exporter/templates/poddisruptionbudget.yaml and its values, CONTRIBUTING.md, .github/workflows/ci.yaml and .github/linters/ct.yaml first. Write templates/poddisruptionbudget.yaml following the blackbox exporter pattern with this chart's own helpers, gated on a podDisruptionBudget value that defaults to an empty map, and copy the value into spec with toYaml. Document the value in the chart README, bump the chart version by a minor step, add a ci values file with two replicas and maxUnavailable 1, and add a helm-unittest file covering the default and the enabled case. Change no other chart. Run helm lint, helm template with and without the ci values, helm unittest --strict and ct lint, and quote the outputs in the pull request.",
"resources": [
{
"url": "https://github.com/your-org/your-app"
}
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
What comes back is a pull request on its own branch touching one chart: a new templates/poddisruptionbudget.yaml, a podDisruptionBudget: {} entry in values.yaml, a row in README.md, Chart.yaml at 6.36.0, a new ci/pdb-values.yaml and a new unittests/poddisruptionbudget_test.yaml. The description explains the opt-in default, shows the rendered budget from the ci values, and points at the blackbox exporter template as the pattern it followed.
The proof is the sandbox output quoted in the description and kept in the run's record: helm lint clean, the two helm template runs, helm unittest --strict passing both cases, and ct lint accepting the version bump. The repository's own ci.yaml runs on the pull request too, installing the chart with the default values and with ci/pdb-values.yaml, which is the first time the budget is created on a real cluster.
Before merging, a reviewer checks that the selector in the rendered budget matches the Deployment's selector exactly, that the README row matches the house format, that the version bump is the right size for the chart's release notes, and that the example in values.yaml does not suggest maxUnavailable: 0 for a single replica. Then they approve and let the release workflow publish the chart.
Run it on every chart with a Deployment and no budget
A team that maintains many charts sends one POST /o/{org}/sessions per chart, with this prompt naming the chart directory and the repository in resources. Because this repository's CI fails a pull request that changes more than one chart, one run per chart is the right shape: a script can list charts/*/templates, grep each for PodDisruptionBudget, and start a run for every chart with a Deployment and no budget. Each run bumps its own chart's version and opens its own pull request.
Set a webhook URL on the API key and Rush POSTs run.completed or run.failed when a run's turn finishes, with session_url so your server can collect the pull request links. Send a fresh UUID as request_id with each request so a retry returns the run already started.
The use case this job follows, with its automation and trigger: Hand off a long task.
Why run it on Rush?
A disruption budget is ten lines of template and a dozen rules about everything around it: the helper names, the version bump, the README table, the one-chart-per-pull-request rule, the values file chart-testing picks up, the unit test plugin. On Rush the agent reads those rules from the repository's own CI config and sibling charts, then runs the same helm lint, helm template, helm unittest and ct lint in its own sandbox before pushing, so the pull request arrives already shaped the way the maintainers' CI expects.
Every command and rendered manifest stay in the run's record, the diff is confined to one chart, and nothing publishes until a person merges and the release workflow runs. The same request can go to every chart in the repository that lacks a budget, one sandbox and one pull request each, without a maintainer copying a template forty times.
Frequently asked questions
Why is the budget off by default?
Because the chart ships replicaCount: 1, and a budget with maxUnavailable: 0 on a single replica would block every node drain. Following the blackbox exporter chart, podDisruptionBudget defaults to an empty map, the template renders nothing, and a user who runs two or more replicas opts in by setting maxUnavailable or minAvailable.
Why copy the value with toYaml instead of fixed fields?
It is the pattern the repository already uses, and it lets a user set either minAvailable or maxUnavailable, as a number or a percentage, plus unhealthyPodEvictionPolicy on newer clusters, without the chart needing a release for each field. The unit test pins the shape so a typo in the value fails fast.
Does it change other charts that lack a budget?
No. The repository's ci.yaml fails a pull request that changes more than one chart, so the agent changes exactly one. Start one run per chart to cover the others; each run bumps that chart's own version.
Which version bump does it make?
A minor one, 6.35.0 to 6.36.0, because an optional new template is a backwards-compatible feature under the semver rules in CONTRIBUTING.md. If your chart tracks another convention, say so in the prompt and the agent follows it.