The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes source code and git metadata, which represents a surface for indirect prompt injection if malicious instructions are embedded within the codebase being reviewed.
Ingestion points: Project files and git diffs are read during 'Phase 1: 초기 분석' and 'Phase 2: 반복 개선 루프' as described in SKILL.md.
Boundary markers: Absent. The instructions do not define specific delimiters to separate untrusted code data from the agent's internal instructions.
Capability inventory: SKILL.md documents the ability to execute shell commands for testing (e.g., pytest, npm test, cargo test) and perform git operations like pushing to remote branches.
Sanitization: Absent. There is no mention of sanitizing or escaping the content of the analyzed files before they are provided to sub-agents.
[COMMAND_EXECUTION]: The skill includes automated logic to detect and execute various testing frameworks and shell commands within the local environment.
Evidence: The '테스트 프레임워크 탐지' section in SKILL.md lists specific commands for execution, including pytest, npm test, cargo test, go test, and make test.

iterative-code-review