The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill is designed to create agents that ingest untrusted external data, posing a significant security risk based on the suggested capabilities.
Ingestion points: As seen in examples/agent-creation-prompt.md, the generated agents (e.g., code-quality-reviewer) are designed to read code changes and PR data which are attacker-controllable.
Boundary markers: The templates provided in references/agent-creation-system-prompt.md do not include any instructions for the generated agents to use delimiters or ignore instructions embedded within the data they process.
Capability inventory: The examples explicitly suggest granting these agents Write and Edit tools while they are processing untrusted input, which could lead to unauthorized code modifications if an injection occurs.
Sanitization: No guidance is provided on sanitizing or validating external content before it is processed by the generated agents.
Privilege Escalation (INFO): The reference file references/permission-modes-rules.md documents the bypassPermissions mode. While this is educational material, it describes a high-risk configuration that removes all system-level safety checks, which could be dangerous if implemented in the generated agents without extreme caution.

agent-development