The Agent Skills Directory

[COMMAND_EXECUTION]: The skill assembles shell commands using user-provided input. Specifically, the fixed-point parameter (branch, tag, or SHA) is interpolated directly into git diff, git log, and git rev-parse commands. This creates a command injection vulnerability if the input is not sanitized by the platform.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection from repository content. Evidence chain: (1) Ingestion points include git commit messages and specification files found in documentation or scratchpad directories. (2) Boundary markers are absent in the prompts sent to sub-agents. (3) Capabilities include shell command execution and sub-agent spawning. (4) Sanitization is absent for the ingested natural language content.

code-review