bilibili-video-helper

Warn

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill explicitly instructs users to provide their full Bilibili browser session cookie ("复制 cookie: ... 整段内容发给我"). Session cookies are highly sensitive and could allow for account takeover if the agent logs the prompt or if the data is leaked.
  • [COMMAND_EXECUTION]: The skill relies on shell commands (curl, yt-dlp, psql) that are constructed using variables derived from external Bilibili API responses. The SQL persistence section specifically builds queries using untrusted data like video titles and content without sanitization, posing a classic SQL injection risk.
  • [DATA_EXFILTRATION]: The skill interacts with a user-provided database via the DATABASE_URL environment variable. When combined with the ability to fetch and process untrusted external content, this creates a potential pathway for data manipulation or unauthorized data transfer.
  • [PROMPT_INJECTION]: The skill ingests untrusted user-generated content, such as Bilibili video comments and AI-generated subtitles, for analysis. It lacks explicit boundary markers or sanitization to prevent these external strings from containing indirect prompt injections that could influence the agent's behavior.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 13, 2026, 12:11 PM