The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data, such as database migrations and security rules, which creates a potential surface for indirect prompt injection. However, its implementation as a defensive auditor with integrated redaction mechanisms minimizes these risks.
Ingestion points: The skill analyzes user-provided repository artifacts including SQL migrations, Firebase security rules (*.rules), application environment configurations, and source code.
Boundary markers: The audit is guided by specific regex signatures rather than free-text interpretation, though no explicit delimiters are provided to the agent to distinguish audited content from instructions.
Capability inventory: Utilizes standard search utilities (ripgrep, git grep) and network probing tools (curl) to verify the security posture of local code and remote endpoints.
Sanitization: Implements a dedicated shell function (red()) using sed to automatically redact sensitive patterns, such as API keys and connection strings, ensuring that discovered credentials are not exposed in the audit findings.
[COMMAND_EXECUTION]: The skill uses command-line tools like ripgrep and curl for their intended purposes of searching codebases and performing network status checks. These commands are configured for read-only inspections and existence probes on authorized targets, aligned with best practices for security auditing.

aegis-baas-platform