The Agent Skills Directory

[SAFE]: The skill does not contain any malicious code, external dependencies, or network-based operations. It serves as a pure instruction-based persona and strategic framework for the agent.
[PROMPT_INJECTION]: The skill includes instructions to adopt the persona of Sunzi. These instructions are well-structured within the SKILL.md file, defining specific boundaries, role-play rules, and workflows. They do not attempt to bypass safety filters or override system-level constraints.
[DATA_EXFILTRATION]: There is no evidence of hardcoded credentials, access to sensitive local file paths (e.g., .ssh, .env), or network commands (curl, wget, fetch) that could be used for data theft.
[REMOTE_CODE_EXECUTION]: The skill does not perform any remote package installations or execute scripts from external URLs. The npx command in the README.md is a manual installation instruction for the user and is not executed by the skill itself at runtime.
[COMMAND_EXECUTION]: The skill defines a text-based reasoning workflow and does not invoke shell commands, subprocesses, or system utilities.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user input and references local markdown files (research.md, demo-conversation.md). While user input is an ingestion point, the skill lacks actionable capabilities (like network access or file writing) that could be exploited via indirect injection. The instructions include a 'Honesty Boundary' to limit the scope of analysis.

sunzi-perspective