The Agent Skills Directory

[PROMPT_INJECTION]: Potential for indirect prompt injection via the analysis of untrusted repository metadata.
Ingestion points: The scripts/analyze_deployment_risk.py script reads and processes the content of README.md from the current repository.
Boundary markers: The script's output does not use delimiters or provide instructions to the agent to treat the ingested content as untrusted data.
Capability inventory: The agent uses the script's output to assess deployment risks and propose infrastructure modifications, which could be influenced by malicious instructions in the README.
Sanitization: The script performs simple string matching on the README content after lowercase conversion, providing no protection against embedded instructions.
[CREDENTIALS_UNSAFE]: Inclusion of example credentials and secrets in reference documentation templates.
Evidence: references/kubernetes.md contains hardcoded example values for DATABASE_URL and API_KEY.
Evidence: references/docker.md includes a DATABASE_URL with a default devpassword.
[EXTERNAL_DOWNLOADS]: Reference to external GitHub Actions from well-known sources without version pinning.
Evidence: references/cicd-pipeline.md references the master branch for aquasecurity/trivy-action.
Context: While referencing a well-known security provider, the use of unpinned versions is a supply chain best practice violation.

engineering-devops-automator