The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill directs the agent to analyze user-provided source code and synthesized CloudFormation templates. This process introduces a surface where malicious instructions embedded in the codebase (such as in code comments or data structures) could influence the agent's logic or decisions during the security review.
Ingestion points: Local source files and synthesis output (cdk.out) are ingested using search tools like rg.
Boundary markers: The instructions do not mandate the use of delimiters or specific safety markers to differentiate between the agent's instructions and the untrusted data from the audited project.
Capability inventory: The skill possesses the ability to execute shell commands (rg, npx cdk, npm), which increases the potential impact of a successful injection.
Sanitization: No explicit sanitization or filtering of the content retrieved from the project codebase is performed before processing.
[COMMAND_EXECUTION]: The workflow relies on executing several shell commands, including ripgrep for pattern discovery and npx cdk synth, npm run build, and npm test for environment validation. These commands represent a capability to execute processes on the host system as part of the normal development and auditing lifecycle.

api-gateway-authorizer-security