Know the impact
before you execute.
propt wraps kubectl transparently
and analyzes live Kubernetes state before mutating commands
are allowed to execute.
Kubernetes operations,
with context.
propt doesn't replace kubectl. It sits in front of it, analyzes the proposed mutation, explains the potential consequences, and leaves the final decision to the engineer.
propt kubectl delete pod payment-api-85648675fb-6bj46
# propt checks:
deployment → replica availability
ownerReferences → resource relationships
selectors → service relationships
Popeye → operational findings
metrics-server → resource utilization
Built around
real cluster state.
No giant command-to-risk lookup table. No invented relationships. No opaque risk score.
Live dependency graph
Builds relationships from actual Kubernetes ownerReferences and label-selector matching. The graph reflects the cluster that exists now.
Impact inference
Combines resource relationships, operational findings and optional resource metrics to explain what a mutation could affect.
Transparent kubectl
Read-only commands pass directly through. Mutating commands are analyzed before the original kubectl command is executed.
Human in the loop
propt does not silently change infrastructure. It presents evidence and asks the engineer whether the command should proceed.
Auditable decisions
Every proposed change, evidence, consequence and execution decision can be recorded as structured JSON in Amazon S3.
Graceful degradation
Missing optional signals are reported instead of fabricated. AWS audit failures never block the underlying kubectl workflow.
One command.
Multiple signals.
propt keeps the analysis pipeline explicit and inspectable from command parsing to final execution.
kubectl command │ ▼ Common Change Model │ ▼ Live Resource Graph │ ├────────────── ownerReferences │ ├────────────── label selectors │ ▼ Impact Inference Engine │ ├────────────── cluster state ├────────────── Popeye findings └────────────── metrics-server │ ▼ Evidence-based Explanation │ ▼ Execute? [y/N] │ ├──────────► YES ──► original kubectl command │ └──────────► NO ──► command cancelled │ ▼ AWS S3 Audit Record
Real signals.
Not guesses.
propt uses existing Kubernetes and operational tooling rather than creating a fictional model of your cluster.
Python client for live resource discovery.
Dependency graph construction and traversal.
Operational cluster findings.
Optional real CPU utilization data.
Original Kubernetes command interface.
AWS S3 audit integration.
Queryable decision history.
Core implementation and CLI tooling.
Infrastructure decisions
shouldn't disappear.
Every analyzed mutation can produce a structured audit record containing the command, target, evidence, consequence and engineer decision.
{ "command": "delete pod payment-api-85648675fb-6bj46", "target": "pod/payment-api-85648675fb-6bj46", "impact_level": "HIGH", "evidence": [ "available replicas below desired" ], "executed": false }
What propt
deliberately doesn't do.
The hackathon scope stays intentionally focused.
No hardcoded risk table
Risk is derived from live cluster state and operational signals rather than a static command-to-risk mapping.
No fabricated metrics
If metrics-server isn't available, propt says so instead of inventing a utilization number.
No autonomous remediation
propt explains the potential impact and lets the engineer decide whether to continue.
Make the decision
before the mutation.
Explore the implementation, run the demo, and see how propt turns Kubernetes operations into evidence-based decisions.