The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (source code and git diffs) through ingestion points in SKILL.md and rules/change-detection.md. It has access to powerful capabilities including shell execution via git, npm, node, and python3. The instructions lack explicit delimiters or boundary markers to prevent malicious code under review from attempting to hijack the agent's instructions, though the skill's primary purpose is to identify such risks.
[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands for project introspection, such as git log, git diff, and npm view. These are standard and necessary for its stated purpose as a code review and release management tool.
[EXTERNAL_DOWNLOADS]: Fetches package metadata from PyPI's official domain (pypi.org) using curl to validate versions. This targets a well-known service for its intended functionality and does not contribute to risk escalation.

ba-guan