The Agent Skills Directory

[PROMPT_INJECTION]: The skill identifies and processes user-provided artifacts (code, design, or documentation) which are then interpolated into prompts for several parallel reviewer agents. This architectural pattern exposes the system to indirect prompt injection where adversarial content within the reviewed artifacts could attempt to manipulate the sub-agents' instructions.
Ingestion points: Artifacts are loaded via the working_scope.artifact parameter defined in SKILL.md.
Boundary markers: The skill utilizes variable placeholders like {conversation_context} and {scope_description} to delineate user content within agent prompts.
Capability inventory: The skill and its sub-agents have access to Read, Write, Bash, and Task tools.
Sanitization: The provided instructions do not specify explicit sanitization or filtering logic for the input artifacts prior to their use in agent prompts.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for users to clone additional required skill modules from the author's GitHub repository at github.com/mikeng-io/agent-skills. This is documented as a standard method for dependency resolution within the vendor's ecosystem.
[COMMAND_EXECUTION]: The skill makes use of restricted shell commands to manage its environment and outputs.
It uses Bash(ls *) and Bash(mkdir *) to verify dependencies and organize report artifacts in the .outputs/review/ directory.
It specifies the use of Bash(git *) for the purpose of downloading and managing necessary skill dependencies from the author's repository.

deep-review