The Agent Skills Directory

[DATA_EXFILTRATION]: The skill transmits source code diffs and requirement documents to external AI providers via the external_review tool. This exposes sensitive project data to third-party systems as part of its core functionality.
[COMMAND_EXECUTION]: The skill executes shell-based git commands (such as git rev-parse and git diff) using variable interpolation. This pattern presents a potential risk if the repository metadata or environmental variables are manipulated by an attacker to include unintended shell sequences.
[PROMPT_INJECTION]: The reviewer subagent processes untrusted content from git diff output. This is a vector for indirect prompt injection, as malicious instructions embedded in code comments or strings could influence the agent's behavior or verdict. The current implementation (in code-reviewer.md) lacks explicit boundary markers or sanitization for this external data, which is processed by a general-purpose subagent with broad tool access.

code-review