argocd-advanced

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides numerous instructions and template scripts (such as bootstrap-script.sh and Workflows/DiagnoseSyncFailure.md) that utilize standard system and DevOps CLI tools.
  • Evidence: Execution of argocd, kubectl, git, yq, az, and bash commands for cluster management, manifest manipulation, and diagnostic tracing.
  • Context: These tools are fundamental to the skill's purpose of managing Kubernetes infrastructure via GitOps and are used within expected operational contexts.
  • [INDIRECT_PROMPT_INJECTION]: The ApplicationSet patterns described in References/applicationset.md define an attack surface where untrusted data from external Git repositories or SCM providers is ingested into the agent's context.
  • Ingestion points: Git directory/file generators, SCM Provider generators, and Pull Request generators which read metadata like labels, branch names, and repository topics.
  • Boundary markers: The skill relies on ArgoCD's internal boundaries and Helm/Kustomize templating; it does not specify explicit agent-level delimiters in its workflows.
  • Capability inventory: Shell command execution (kubectl, argocd, git) and local file system write access.
  • Sanitization: The skill mandates the use of yq eval, kustomize build, and pre-commit validation hooks to verify manifests before they are committed or applied.
  • [SAFE]: The skill implements a local notification mechanism using HTTP POST requests to localhost:8888 for status reporting.
  • Evidence: curl -s -X POST http://localhost:8888/notify is used in multiple workflows to provide voice or status updates to the local development environment.
  • Context: This is documented as part of the organizational notification system and targets a loopback address, posing no external exfiltration risk.
  • [SAFE]: The skill provides excellent guidance on secret management, ensuring sensitive credentials are not exposed.
  • Evidence: Templates for cluster secrets and registry authentication (e.g., Samples/image-updater/authentication-secrets.yaml) use clear placeholders and include explicit security warnings against committing real credentials to version control.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 06:02 PM