social-media-publishing

Warn

Audited by Gen Agent Trust Hub on Aug 31, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXPOSURE]: The skill scripts access sensitive credential and token files stored in the user's home directory (e.g., ~/.secrets/linkedin/env, ~/.secrets/reddit/env, ~/.secrets/x/env, ~/.secrets/linkedin/posting.tokens.json). While this is the intended mechanism for authentication, it involves programmatic access to private secret stores.
  • [COMMAND_EXECUTION]: The script scripts/reddit/native_video.py executes shell commands using subprocess.run to call ffmpeg for video encoding. While input paths are managed by the tool, shell execution on media files presents an attack surface if malicious files are processed.
  • [DYNAMIC_EXECUTION]: The script scripts/youtube/cli.py uses os.execve to replace the current process with a new one using a different Python interpreter (from a virtual environment). This is used to ensure the modal library is available but constitutes a high-privilege process operation.
  • [DYNAMIC_EXECUTION]: The script scripts/reddit/praw_client.py uses importlib.import_module to dynamically load the praw library at runtime.
  • [EXTERNAL_DOWNLOADS]: Multiple platform CLIs (LinkedIn, X, YouTube, Instagram, TikTok) support a --video-url or --media-url flag that downloads media from arbitrary remote HTTPS URLs to a local temporary directory using urllib.request. This data is then processed (and in some cases encoded) before being uploaded to social media APIs.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing untrusted data from various sources.
  • Ingestion points: CLI arguments (--text, --title), local Markdown files (--text-file, --description-file, --comment-file), and JSON plan files (--plan).
  • Boundary markers: None detected in the prompt interpolation logic.
  • Capability inventory: Network exfiltration to major social media APIs (Reddit, LinkedIn, X, YouTube), shell command execution (ffmpeg), and local file system writes.
  • Sanitization: Minimal sanitization is performed beyond basic string stripping.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 31, 2026, 05:40 AM
Security Audit — agent-trust-hub — social-media-publishing