The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted sources such as URLs and PDF documents. This creates a surface for Indirect Prompt Injection.
[Ingestion Points]: Reference materials provided as URLs, PDFs, or external text snippets (defined in SKILL.md under the 'Receiving Information' section).
[Boundary Markers]: The skill includes a dedicated 'Security Requirements' section that explicitly instructs the agent to treat third-party content as 'factual material' rather than 'instructional sources' and to ignore any embedded commands or role-play attempts.
[Capability Inventory]: The skill has the capability to write files (content/post/<slug>/index.md), perform network checks (curl), and execute local scripts via node.
[Sanitization]: While the skill provides high-quality instructional guidelines for the agent to ignore malicious text, it does not specify programmatic sanitization or escaping for the shell commands it generates.
[COMMAND_EXECUTION]: The workflow involves executing shell commands to generate cover images using a Node.js script.
[Evidence]: Step 4 in SKILL.md instructs the agent to run: node "<generate-cover 路径>/scripts/index.js" -t "文章标题" -s "文章摘要" -l "标签" -a "作者" ....
[Risk]: The arguments for these commands (title, summary, tags) are derived from the external reference materials. If an attacker controls the content of a summarized URL, they could attempt to include shell metacharacters (e.g., backticks, semicolons, or subshells) in the title/summary to achieve command injection during the node execution phase.
[EXTERNAL_DOWNLOADS]: The skill performs a network connectivity check using curl against https://registry.npmjs.org.
[Context]: This targets a well-known public registry to verify internet access before attempting cover generation. Under the [TRUST-SCOPE-RULE], this is considered a safe operation for environment validation.

content-creator