The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard shell commands (find, ls, cat) and cloud CLI tools (aws sts get-caller-identity, gcloud config get-value project) to identify the local environment and locate configuration files. This behavior is transparent and aligned with the tool's primary purpose of infrastructure discovery.
[DATA_EXPOSURE]: The skill reads sensitive infrastructure files, including Terraform scripts, Kubernetes manifests, and cloud configuration files (wrangler.toml, fly.toml). This access is necessary to perform the security audit, which specifically looks for risks like hardcoded secrets and overly permissive IAM roles.
[PROMPT_INJECTION]: The skill contains potential vulnerability to indirect prompt injection as it ingests and processes untrusted data from the user's project files.
Ingestion points: Files discovered via find . -name '*.tf', ls k8s/, etc. in Step 0.
Boundary markers: Not explicitly defined in the prompt interpolation; the agent is simply instructed to read the files.
Capability inventory: The agent has access to Bash, Read, and WebFetch tools (as per SKILL.md frontmatter).
Sanitization: No specific sanitization or filtering of the file content is described before analysis.

forge-audit