The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is designed to ingest and process external content (codebases, architectural descriptions, and existing documentation) to generate output. This creates a significant attack surface where malicious instructions hidden in the data being processed could be executed by the agent.
Ingestion Points: User requests and project files within the workspace.
Boundary Markers: None specified in the SKILL.md.
Capability Inventory: Includes Bash, Write, Edit, and Grep which allow for significant system modification.
Sanitization: No sanitization or validation mechanisms are described for handled content.
[Command Execution] (HIGH): The inclusion of the Bash tool in allowed-tools provides the capability to execute arbitrary shell commands. When combined with the documentation-generation purpose, there is a risk that a prompt injection could trigger dangerous shell commands.
[Data Exposure] (MEDIUM): With Read, Write, and Edit permissions, the skill has broad access to the user's workspace. If compromised via injection, sensitive project files or environment variables could be accessed or modified.

architecture-doc-creator