The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze and generate CI/CD manifests (e.g., Dockerfiles, GitHub Action workflows) which may contain instructions from untrusted sources in a real-world project environment.
Ingestion points: Project manifest files (e.g., Dockerfile, .github/workflows/*.yml) processed in SKILL.md.
Boundary markers: None explicitly mentioned to separate untrusted manifest content from agent instructions.
Capability inventory: File writing, subprocess execution (docker, gcloud, vercel, go), and network operations for deployment (SKILL.md).
Sanitization: No specific sanitization or validation logic is defined for the content of processed manifests.
[REMOTE_CODE_EXECUTION]: The skill recommends executing go run golang.org/x/vuln/cmd/govulncheck@latest, which involves fetching and running code from a remote repository. This is a standard security auditing practice using a trusted domain (golang.org) and does not represent a malicious pattern.
[DATA_EXFILTRATION]: The skill actively encourages the use of secret scanning tools (trufflehog) and warns against including credentials or sensitive files like .env and .git in container images, demonstrating a strong security posture.

ci-cd