The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is instructed to search for and read sensitive configuration files, specifically checking .env*, docker-compose, and sample environment files to identify telemetry service credentials and API keys.
[DATA_EXFILTRATION]: The assessment workflow includes a step to probe external telemetry backends. If credentials are found in the codebase, the agent is directed to perform network queries and fetch data samples from these external services (e.g., Langfuse, LangSmith, or other OTLP backends).
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform codebase analysis (using grep, jq, and cat). It also executes a bundled Python script (assets/render_report.py) and utilizes platform-specific commands (open, xdg-open, or start) to automatically launch the generated HTML report in a browser.
[PROMPT_INJECTION]: The skill functions by ingesting and analyzing external, untrusted codebase data, including prompt templates and tool definitions. This represents a surface for indirect prompt injection, where malicious instructions hidden in the partner's codebase could potentially influence the agent's behavior or the outcome of the security report.
Ingestion points: Codebase manifests, source files, and prompt templates (Step 1, Step 2, and Step 4).
Boundary markers: None explicitly implemented in the commands used to read file contents.
Capability inventory: Access to Bash (read/write), Agent (delegation), and network capabilities.
Sanitization: No explicit sanitization or filtering is described for the content ingested from the analyzed repository.

ratel-assessment