The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute specific Python scripts provided in the writing-rules.md file (specifically under sections [2026-04-28] and [2026-05-22]). These scripts use regular expressions to modify files on the local filesystem. Instructing an agent to run arbitrary code embedded in skill documentation is a form of dynamic execution that can be misused if the script logic is altered or if the agent is directed to sensitive file paths.
[DATA_EXFILTRATION]: During the '事实核查' (Fact Checking) stage, the skill utilizes WebSearch to verify content from the user's draft. This process inherently transmits segments of the user's potentially private data to third-party search engines. While this is necessary for the skill's stated purpose, it constitutes data transmission to external domains.
[PROMPT_INJECTION]: The skill uses WebFetch to ingest information from the internet. This creates a surface for indirect prompt injection, as an attacker-controlled website could include hidden instructions to influence the agent's writing or behavior. The skill lacks explicit boundary markers or sanitization for data retrieved via these tools.
[SAFE]: The skill accesses local configuration files within .claude/skills/write-article/ and reads from the project's README.md. These operations are well-scoped to the skill's primary function as an article management tool and do not involve unauthorized system access.

write-article