The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides utility scripts that are susceptible to command injection if their arguments are sourced from untrusted external data without validation.
Ingestion points: Scripts such as collect-evidence.sh in references/incident-response.md and release.sh in references/release-automation.md accept command-line arguments (e.g., INCIDENT_ID, VERSION) that may be derived from external events.
Boundary markers: There are no instructions or boundary markers defined to ensure the agent validates or sanitizes these inputs before use.
Capability inventory: The skill provides access to powerful tools including kubectl exec, gh (GitHub CLI), and git, which increase the potential impact of an injection attack.
Sanitization: Inputs are interpolated directly into shell strings (e.g., mkdir -p incidents/${INCIDENT_ID}/evidence) without escaping or validation.
[COMMAND_EXECUTION]: The skill includes several scripts designed to automate infrastructure and incident response tasks.
Evidence: references/incident-response.md contains scripts that execute kubectl exec, tcpdump, and psql to collect forensics. references/release-automation.md includes a coordination script that uses gh and git for release management.
[EXTERNAL_DOWNLOADS]: The skill references and utilizes various well-known third-party tools and GitHub Actions for security scanning and deployment.
Evidence: The skill utilizes aquasecurity/trivy-action for container scanning, sigstore/cosign-installer for artifact signing, and official actions from the actions/ and docker/ GitHub organizations. These are recognized as well-known and reputable services.

devops-engineer