The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages the bash tool to execute the codex and copilot CLI utilities. It includes safety checks for tool availability and utilizes security-oriented flags, such as --sandbox read-only for Codex and --no-ask-user for Copilot, to restrict the tools' execution environment.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted code from a user-defined target path. Malicious instructions embedded in the reviewed code could potentially influence the behavior of the review agents.
Ingestion points: External source code provided in the <target> argument (SKILL.md).
Boundary markers: None detected; code content is processed directly within the review prompts without explicit delimiters.
Capability inventory: Utilizes Task() for sub-agent spawning and bash for external CLI execution (SKILL.md).
Sanitization: No specific sanitization or escaping of the target code is performed prior to processing.
[SAFE]: The overall design of the skill is transparent and its capabilities are well-aligned with its described purpose of performing non-destructive code analysis. The installation methods and external references are consistent with the author's established ecosystem.

multi-review