The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs users to store highly sensitive authentication data, including session cookies and API signatures (x-ca-key, x-ca-nonce, x-ca-signature), in a local plain-text JSON file (csdn_config.json). Session cookies are particularly sensitive as they represent a full authenticated session.
[DATA_EXFILTRATION]: The core script (scripts/csdn_article.js) reads the sensitive session cookies and signatures from the local config file and transmits them via POST request to the CSDN API endpoint (https://bizapi.csdn.net/blog-console-api/v3/mdeditor/saveArticle). While this is the intended functionality, the storage and handling of session-level credentials create a high-risk data path.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by reading and processing local Markdown files from the csdnarticle/ directory. These files, which may be generated from external sources, are parsed and sent to the CSDN API without explicit boundary markers or sanitization against embedded agent instructions.
Ingestion points: Reads content from local Markdown files via the --file argument in scripts/csdn_article.js.
Boundary markers: Absent; the content is processed directly.
Capability inventory: Involved in network requests (https.request) and file system access (fs.readFileSync, fs.writeFileSync).
Sanitization: Uses the marked library to convert Markdown to HTML, but does not sanitize for agent-specific instruction markers.
[COMMAND_EXECUTION]: The skill relies on the execution of a Node.js script that performs file system writes (creating csdnarticle/ and csdn_article_map.json) and network operations based on command-line arguments, which could be exploited if the agent is tricked into passing malicious parameters.

csdn-article-publish