yt-retention

Warn

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script retention.py dynamically modifies the Python search path to import a module (deadair) from a sibling directory (../yt-edit).
  • Evidence: sys.path.insert(0, os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "yt-edit")) in retention.py.
  • Risk: This allows execution of code from a path computed at runtime. This is an unverifiable dependency that could lead to the execution of malicious code if the local environment or sibling directories are compromised.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from YouTube retention CSVs and SRT transcript files, which are then reflected into the agent's context.
  • Ingestion points: retention.csv and transcript.srt (specified via the --transcript argument) are read in retention.py.
  • Boundary markers: Absent; the extracted text from the transcript is printed directly into the output consumed by the agent without delimiters.
  • Capability inventory: The retention.py script has file-read capabilities (CSV/SRT) and prints to stdout. While the script itself lacks network/shell capabilities, the agent interpreting the output may have such access.
  • Sanitization: Absent; the script truncates transcript snippets to 140 characters but does not sanitize or escape the text to prevent embedded prompt injection attacks.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 9, 2026, 12:32 PM