The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's documentation recommends that users install the gateguard-ai package via pip. This package is not provided by a recognized trusted vendor or a well-known service. Furthermore, it does not match the established naming patterns for the skill's author ('affaan-m'), posing a potential supply chain risk if the package were to be compromised or malicious.
[INDIRECT_PROMPT_INJECTION]: The core mechanism of the skill requires the agent to ingest and process untrusted data to satisfy the "fact-forcing" gate. This creates a surface for indirect prompt injection attacks where malicious content in the environment could manipulate the agent's behavior.
Ingestion points: The agent is instructed to read and process file imports, data schemas, and verbatim user instructions from the project environment (SKILL.md).
Boundary markers: There are no explicit boundary markers or instructions to treat the quoted user instructions as untrusted data, increasing the risk that embedded malicious commands could be executed.
Capability inventory: The skill facilitates access to powerful tools including bash, Edit, Write, and MultiEdit (SKILL.md).
Sanitization: While the skill suggests using redacted or synthetic values for data schemas, it provides no sanitization or escaping guidance for other ingested data, such as file imports or user instruction quotes.
[COMMAND_EXECUTION]: The skill explicitly directs the agent to use shell commands like grep, bash, and find to investigate the file tree before performing edits. Frequent execution of these commands on a codebase containing untrusted content increases the risk of command injection if the agent handles filenames or file content with shell metacharacters unsafely.

gateguard