The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from git diffs, which serves as an untrusted data source. An attacker could place malicious instructions within code comments or strings in a Pull Request to subvert the agent's logic or omit specific findings.\n
Ingestion points: The skill reads a temporary file diff_temp.txt created via git diff (SKILL.md).\n
Boundary markers: There are no explicit instructions for the agent to disregard instructions found within the diff content itself.\n
Capability inventory: The skill performs file I/O and shell command execution (git diff).\n
Sanitization: While it filters out SQL and template noise, it does not include sanitization for embedded natural language instructions.\n- [COMMAND_EXECUTION]: The skill instructs the agent to run git diff commands on the local system to extract code changes (SKILL.md). This interaction with the host shell is a standard but noteworthy capability.

sdlc-config-review