Aks Network Capture
MicrosoftOfficialCreated by Microsoft
- Skill
- Workflow
- Cloud and DevOps
- MIT
- Low risk
- Has scripts
AutoGineer summary
Written by AutoGineer from the skill’s own files, not by the author
The aks-network-capture skill provides a governed procedure for collecting wire-level evidence from Azure Kubernetes Service (AKS) nodes. It is designed for scenarios requiring explicit packet capture after initial read-only diagnostics have been exhausted. The skill includes scripts for setting up capture configurations, executing the capture, generating test traffic, and retrieving the resulting data, alongside tools for collecting relevant Azure network metadata. It emphasizes strict authorization, target binding, and resource cleanup to ensure that captures are performed securely and within defined operational boundaries.
What it does
- Configures AKS nodes for packet capture using dedicated scripts.
- Executes bounded packet captures on specific nodes or pods.
- Collects Azure network configuration metadata for context.
- Generates test traffic to facilitate network troubleshooting.
- Retrieves capture artifacts and manages resource cleanup by run ID.
When to use it
- When you need to perform a packet capture on an AKS node.
- When you need to prove where network packets are being dropped.
- After read-only diagnostics have failed to resolve a connectivity issue.
- When you require wire-level evidence for network troubleshooting.
Advantages
- Provides a structured, repeatable process for network evidence collection.
- Includes built-in authorization and target binding steps to maintain security.
- Supports both Bash and PowerShell environments for flexibility.
- Ensures resource cleanup is tied to specific capture run IDs.
Usage examples
- “Capture packets on an AKS node to investigate a drop.”
- The agent initiates the capture workflow, starting with authorization and target binding, followed by the deployment of the capture configuration.
- “Run tcpdump on AKS to troubleshoot connectivity.”
- The agent guides you through the setup and execution of the packet capture scripts, ensuring the capture is bounded and authorized.
Install
Download the skill as a ZIP:
https://autogineer.com/agent-skills/microsoft/aks-network-capture/downloadUnzip it into the cross-client skills folder of your project:
mkdir -p .agents/skills && unzip aks-network-capture.zip -d .agents/skills
Documentation by Microsoft
AKS Network Capture
Quick Reference
| Use | Requires | Safety |
|---|---|---|
| AKS pcap evidence | kubectl; az for Azure evidence | Bounded, pinned, least privilege |
When to Use This Skill
Use for explicit packet capture after read-only checks, not generic connectivity failures.
MCP Tools
Azure MCP's AKS area provides cluster and node-pool metadata, not Kubernetes command execution or packet capture.
Host Capability Gate
Before executing the workflow, confirm that the host permits the required
Bash or PowerShell execution, kubectl access to the bound cluster, az for
Azure evidence, access to the bundled scripts, and an approved artifact
destination. A governed Azure CLI tool alone does not establish that shell,
Kubernetes commands, or artifact operations are supported.
If a required capability is unavailable or prohibited, state that capture
execution is unavailable in this host. Analyze supplied, appropriately
redacted evidence or give the operator a target-bound collection/capture plan;
do not claim to have run it. Never route kubectl through Azure MCP, add an
unapproved execution path, or bypass host policy. Host support does not replace
the mutation and sensitive-data approvals below.
Run bundled scripts from the skill root only after this gate is satisfied.
Workflow/Steps
Check host capabilities, then complete authorization and target binding. Capture intent is not mutation consent. Stop for separate approval before any debug-container fallback.
Install Bash / PowerShell.
Capture nodes or pods with Bash / PowerShell.
Generate traffic if approved with Bash / PowerShell.
Retrieve the exact run with Bash / PowerShell.
Gather Azure evidence with Bash / PowerShell.
The ConfigMap runs run-capture.sh inside its pinned Linux image.
Error Handling
| Error | Action |
|---|---|
| Invalid input | Correct it before retrying. |
| Missing/stale ConfigMap | Run setup again. |
| Capture/retrieval failure | Inspect Job logs; missing evidence is not success. |
Files
- ATTRIBUTION.mdAdded by AutoGineer983 B
- LICENSELicence file added by AutoGineer1.1 KiB
- SKILL.md2.9 KiB
- references/capture-authorization.md1.4 KiB
- scripts/collect-azure-network-info.ps1Script7.9 KiB
- scripts/collect-azure-network-info.shScript7.0 KiB
- scripts/create-capture.ps1Script11.5 KiB
- scripts/create-capture.shScript12.4 KiB
- scripts/generate-test-traffic.ps1Script5.4 KiB
- scripts/generate-test-traffic.shScript6.1 KiB
- scripts/retrieve-captures.ps1Script9.4 KiB
- scripts/retrieve-captures.shScript8.1 KiB
- scripts/run-capture.shScript2.4 KiB
- scripts/setup-capture-configmap.ps1Script1.7 KiB
- scripts/setup-capture-configmap.shScript1.2 KiB
scripts/retrieve-captures.sh
#!/usr/bin/env bash
# Retrieve exact-run capture bundles from nodes and clean up capture resources.
# Exit codes: 0 = success/help, 1 = retrieval/cleanup failure, 2 = usage/argument error.
set -euo pipefail
CAPTURE_NAME=""
OUTPUT_PATH="/var/log/aks-network-captures"
WORKSPACE_DIR="${WORKSPACE_DIR:-./aks-network-captures}"
NAMESPACE="default"
RETRIEVE_RUN_ID=""
usage() {
cat <<EOF
Usage: $0 --name <capture-name> [options]
Required:
--name <string> Capture name used during creation
Options:
--output-path <path> Node hostPath base (default: ${OUTPUT_PATH})
--workspace-dir <path> Local workspace directory (default: ${WORKSPACE_DIR})
--namespace <string> Capture Job namespace (default: default)
--run-id <string> Exact run ID; required when multiple runs exist
-h, --help Show this help
EOF
}
usage_error() {
echo "Error: $*" >&2
usage >&2
exit 2
}
die() {
echo "Error: $*" >&2
exit 1
}
require_value() {
[ "$#" -ge 2 ] && [ -n "$2" ] || usage_error "$1 requires a value"
}
valid_single_line() {
case "$1" in
*$'\r'*|*$'\n'*) return 1 ;;
esac
}
valid_rfc1123() {
valid_single_line "$1" && [ "${#1}" -le 63 ] \
&& printf '%s' "$1" | grep -Eq '^[a-z0-9]([-a-z0-9]*[a-z0-9])?$'
}
valid_node_name() {
valid_single_line "$1" && [ "${#1}" -le 253 ] \
&& printf '%s' "$1" \
| grep -Eq '^[a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*$'
}
valid_hostpath() {
case "$1" in
/*) ;;
*) return 1 ;;
esac
case "$1" in
*..*|*[!A-Za-z0-9/_.-]*) return 1 ;;
esac
}
valid_run_id() {
local run_date="${1%%-*}"
local remainder="${1#*-}"
local run_time="${remainder%%-*}"
local run_token="${remainder#*-}"
case "$1" in
*[!0-9a-f-]*) return 1 ;;
esac
[ "${#run_date}" -eq 8 ] && [ "${#run_time}" -eq 6 ] \
&& [ "${#run_token}" -eq 24 ] || return 1
case "${run_date}${run_time}${run_token}" in
*[!0-9a-f]*) return 1 ;;
esac
}
while [ "$#" -gt 0 ]; do
case "$1" in
--name)
require_value "$@"; CAPTURE_NAME="$2"; shift 2 ;;
--output-path)
require_value "$@"; OUTPUT_PATH="$2"; shift 2 ;;
--workspace-dir)
require_value "$@"; WORKSPACE_DIR="$2"; shift 2 ;;
--namespace)
require_value "$@"; NAMESPACE="$2"; shift 2 ;;
--run-id)
require_value "$@"; RETRIEVE_RUN_ID="$2"; shift 2 ;;
-h|--help)
usage; exit 0 ;;
*)
usage_error "unknown option: $1" ;;
esac
done
[ -n "$CAPTURE_NAME" ] || usage_error "--name is required"
valid_rfc1123 "$CAPTURE_NAME" || usage_error "--name must be an RFC 1123 label"
valid_hostpath "$OUTPUT_PATH" \
|| usage_error "--output-path must be absolute and contain only letters, digits, '/', '_', '.', or '-'"
valid_single_line "$WORKSPACE_DIR" || usage_error "--workspace-dir must be a single line"
valid_rfc1123 "$NAMESPACE" || usage_error "--namespace must be an RFC 1123 label"
[ -z "$RETRIEVE_RUN_ID" ] || valid_run_id "$RETRIEVE_RUN_ID" \
|| usage_error "--run-id has an invalid format"
command -v kubectl >/dev/null 2>&1 || die "kubectl is not installed or not on PATH"
if [ -z "$RETRIEVE_RUN_ID" ]; then
if ! CAPTURE_RUNS="$(kubectl get jobs -n "$NAMESPACE" \
-l "capture-id=${CAPTURE_NAME}" \
-o jsonpath='{range .items[*]}{.metadata.labels.capture-run}{"\n"}{end}' \
| sed '/^$/d' | sort -u)"; then
die "could not list capture runs"
fi
RUN_COUNT="$(printf '%s\n' "$CAPTURE_RUNS" | sed '/^$/d' | wc -l | tr -d '[:space:]')"
case "$RUN_COUNT" in
0) die "no capture runs found for '$CAPTURE_NAME'" ;;
1) RETRIEVE_RUN_ID="$CAPTURE_RUNS" ;;
*) die "multiple capture runs found; specify --run-id" ;;
esac
fi
valid_run_id "$RETRIEVE_RUN_ID" || die "Kubernetes returned an invalid capture run identity"
if ! CAPTURE_JOBS="$(kubectl get jobs -n "$NAMESPACE" \
-l "capture-id=${CAPTURE_NAME},capture-run=${RETRIEVE_RUN_ID}" \
-o jsonpath='{.items[*].metadata.name}')"; then
die "could not list capture Jobs"
fi
[ -n "$CAPTURE_JOBS" ] || die "no Jobs found for the requested capture run"
read -r -a CAPTURE_JOB_NAMES <<<"$CAPTURE_JOBS"
CAPTURES_DIR="${WORKSPACE_DIR}/network-captures"
TIMESTAMP="$(date -u +%Y%m%d-%H%M%S)"
CAPTURE_OUTPUT_DIR="${CAPTURES_DIR}/${CAPTURE_NAME}-${RETRIEVE_RUN_ID}-${TIMESTAMP}"
mkdir -p "$CAPTURE_OUTPUT_DIR"
TEMP_PODS=()
cleanup_resources() {
local failed=0
local pod
if [ "${#TEMP_PODS[@]}" -gt 0 ]; then
for pod in "${TEMP_PODS[@]}"; do
if ! kubectl delete pod "$pod" -n "$NAMESPACE" --ignore-not-found >/dev/null; then
echo "Error: failed to delete retrieval pod '$pod'" >&2
failed=1
fi
done
fi
if ! kubectl delete jobs -n "$NAMESPACE" "${CAPTURE_JOB_NAMES[@]}" \
--ignore-not-found >/dev/null; then
echo "Error: failed to delete capture Jobs" >&2
failed=1
fi
return "$failed"
}
cleanup_on_exit() {
local status=$?
trap - EXIT
if ! cleanup_resources; then
status=1
fi
exit "$status"
}
trap cleanup_on_exit EXIT
for job in "${CAPTURE_JOB_NAMES[@]}"; do
if ! NODE_NAME="$(kubectl get job "$job" -n "$NAMESPACE" \
-o jsonpath='{.spec.template.spec.nodeName}')"; then
die "could not read the node for Job '$job'"
fi
if ! RUN_ID="$(kubectl get job "$job" -n "$NAMESPACE" \
-o jsonpath='{.spec.template.spec.containers[0].env[?(@.name=="RUN_ID")].value}')"; then
die "could not read the run identity for Job '$job'"
fi
valid_node_name "$NODE_NAME" || die "Job '$job' returned an invalid node name"
valid_run_id "$RUN_ID" || die "Job '$job' returned an invalid run identity"
[ "$RUN_ID" = "$RETRIEVE_RUN_ID" ] || die "Job '$job' does not match the requested run"
if ! POD_NAME="$(kubectl get pods -n "$NAMESPACE" -l "job-name=${job}" \
-o jsonpath='{.items[0].metadata.name}')"; then
die "could not find the capture pod for Job '$job'"
fi
[ -n "$POD_NAME" ] || die "no capture pod found for Job '$job'"
if ! POD_STATUS="$(kubectl get pod "$POD_NAME" -n "$NAMESPACE" \
-o jsonpath='{.status.phase}')"; then
die "could not read capture pod status"
fi
[ "$POD_STATUS" = "Succeeded" ] \
|| die "capture pod '$POD_NAME' did not succeed; inspect its logs before retrying"
capture_slug="$(printf '%s' "$CAPTURE_NAME" | cut -c1-20)"
node_slug="$(printf '%s' "$NODE_NAME" | tr '._' '--' | cut -c1-20)"
if ! TEMP_POD_NAME="$(kubectl create -f - -o jsonpath='{.metadata.name}' <<EOF
apiVersion: v1
kind: Pod
metadata:
generateName: retrieve-${capture_slug}-${node_slug}-
namespace: ${NAMESPACE}
labels: { app: aks-network-capture, capture-id: "${CAPTURE_NAME}", role: retrieval }
spec:
hostNetwork: true
nodeName: "${NODE_NAME}"
restartPolicy: Never
containers:
- name: retrieve
image: mcr.microsoft.com/cbl-mariner/busybox:2.0@sha256:e4fb4d51fc9b70d6cdc1ce66a0af02ab40554d2ca632e1d188fabc760e432fdd
command: ["sh", "-c", "sleep 300"]
volumeMounts:
- name: capture-output
mountPath: /capture-root
volumes:
- name: capture-output
hostPath:
path: ${OUTPUT_PATH}/${CAPTURE_NAME}
type: Directory
EOF
)"; then
die "failed to create a retrieval pod on node '$NODE_NAME'"
fi
[ -n "$TEMP_POD_NAME" ] || die "Kubernetes did not return a retrieval pod name"
TEMP_PODS+=("$TEMP_POD_NAME")
kubectl wait -n "$NAMESPACE" --for=condition=Ready \
"pod/${TEMP_POD_NAME}" --timeout=60s
BUNDLE_NAME="capture-${CAPTURE_NAME}-${NODE_NAME}-${RUN_ID}.tar.gz"
RUN_DIR="/capture-root/${RUN_ID}"
kubectl cp -n "$NAMESPACE" "${TEMP_POD_NAME}:${RUN_DIR}/${BUNDLE_NAME}" \
"${CAPTURE_OUTPUT_DIR}/${BUNDLE_NAME}" -c retrieve
[ -s "${CAPTURE_OUTPUT_DIR}/${BUNDLE_NAME}" ] \
|| die "the copied capture bundle '$BUNDLE_NAME' is empty"
# Positional arguments keep validated values out of the fixed remote command.
# shellcheck disable=SC2016
kubectl exec -n "$NAMESPACE" "$TEMP_POD_NAME" -c retrieve -- sh -c \
'rm -f "$1/$2" "$1/$3" && rmdir "$1"' sh \
"$RUN_DIR" "$BUNDLE_NAME" "capture-${CAPTURE_NAME}-${NODE_NAME}-${RUN_ID}.pcap"
kubectl delete pod "$TEMP_POD_NAME" -n "$NAMESPACE" --ignore-not-found
done
cleanup_resources || die "failed to clean up one or more capture resources"
trap - EXIT
echo "Capture bundles saved to: $CAPTURE_OUTPUT_DIR"
ls -lh "$CAPTURE_OUTPUT_DIR"
Requirements and compatibility
Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta
“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.
Needs: Needs Node.js
Security
What our scanner noticed· Low risk
- Scripts that your agent may run
This skill includes scripts that your agent may run. Read them before you use it.
A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.
Attribution and licence
- Created by
- Microsoft
- Publisher
- Microsoft (Official)
- Source
- microsoft/skills · .github/plugins/aks-skills/skills/aks-network-capture · commit 0696600
- Licence
- MIT License (redistribution allowed)
- LICENSE (added by AutoGineer from .github/plugins/aks-skills/LICENSE)
- ATTRIBUTION.md (written by AutoGineer)
- Synced
AutoGineer is not affiliated with or endorsed by Microsoft or Microsoft.
Download ZIP
The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download aks-network-capture.zip