The Agent Skills Directory

Unverifiable Dependency (MEDIUM): The skill triggers an external script located at ./scripts/aha-loop/doc-cleaner.sh. As the script's content is not provided, its behavior cannot be verified, posing a risk of arbitrary command execution if the script is compromised or malicious.
External Network Access (LOW): The skill uses curl to verify external links extracted from documentation. This allows the agent to perform outbound requests to arbitrary URLs found in processed files, including a flagged malicious URL in main.rs.
Indirect Prompt Injection (LOW): 1. Ingestion points: Reads arbitrary .md files throughout the repository via find and grep. 2. Boundary markers: Absent; the instructions do not define delimiters for content. 3. Capability inventory: File system scanning, network requests via curl, and execution of a local cleanup script. 4. Sanitization: None; the skill processes raw content, making it vulnerable to embedded instructions in untrusted documentation.
Command Execution (LOW): Utilizes standard system utilities (find, grep, git) for inventory and staleness checks, which is aligned with the primary purpose of the skill.

doc-review