The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The instructions in SKILL.md include a template for the AI to provide users with a command that downloads and executes a script directly from the internet: curl -fsSL https://opencode.ai/install | bash. Piped remote execution from unverified sources is a major security risk.
[COMMAND_EXECUTION]: The skill explicitly includes shell commands (bash) designed to be executed in the user's environment as part of its 'tutorial' style examples.
[EXTERNAL_DOWNLOADS]: The skill references an external domain https://opencode.ai for script installation, which is not a recognized trusted vendor or well-known service.
[DATA_EXPOSURE]: Under the '参考资料' (References) section in SKILL.md, the skill exposes an absolute local file path: /Users/fred/Documents/GitHub/cycleuser/blog. This reveals private directory structures and user information.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection:
Ingestion points: The /简写 and /write commands in SKILL.md ingest arbitrary user-provided text or topics.
Boundary markers: Absent. The skill does not use delimiters or instructions to ignore embedded commands in the input data.
Capability inventory: While the skill doesn't ship with executable scripts, it provides templates for dangerous shell commands (curl | bash) which could be manipulated if an attacker provides malicious input to the writing commands.
Sanitization: Absent. There is no evidence of input validation or escaping for the processed text.

brief-write