bilibili-source

Fail

Audited by Gen Agent Trust Hub on Jun 13, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/bili-subs.sh executes yt-dlp with the --cookies-from-browser flag. This command accesses local browser sqlite databases to retrieve session cookies required for Bilibili's login-gated subtitle API. While this is the intended functionality for fetching transcripts, it involves sensitive local data access.
  • [COMMAND_EXECUTION]: The script scripts/bili-danmaku.sh uses a multi-line shell pipe to python3 -c to perform raw DEFLATE decompression on API responses. This is a local data transformation step using a static python one-liner.
  • [EXTERNAL_DOWNLOADS]: The skill communicates with Bilibili's public APIs (api.bilibili.com) and resolves short links via b23.tv. These are well-known services and the network operations are restricted to the skill's primary purpose.
  • [REMOTE_CODE_EXECUTION]: Evaluation of the automated scanner alert regarding the pipe from curl to python3 in scripts/bili-danmaku.sh. Analysis confirms this is a false positive for remote code execution; the data from the URL is passed to a local python script for decompression, not executed as a script itself.
  • [PROMPT_INJECTION]: Indirect prompt injection surface detected.
  • Ingestion points: scripts/bili-danmaku.sh reads arbitrary danmaku text from an external API.
  • Boundary markers: Absent; danmaku text is emitted as raw lines.
  • Capability inventory: The skill has access to shell execution via bundled scripts.
  • Sanitization: Absent; no filtering or escaping is applied to the bullet comment text before it is presented to the agent.
Recommendations
  • HIGH: Downloads and executes remote code from: https://api.bilibili.com/x/v1/dm/list.so?oid=$cid - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 13, 2026, 05:03 AM
Security Audit — agent-trust-hub — bilibili-source