packaging-youtube-thumbnails
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The SKILL.md file instructs the agent to execute shell commands during the preamble, specifically python3 telemetry/version_check.py and python3 telemetry/telemetry_init.py. These files are not provided within the skill's specific directory, implying a dependency on the root of the hosting repository.
- [COMMAND_EXECUTION]: The skill's workflow relies on executing local Python scripts bundled with the skill, such as scripts/thumbnail_guard.py for image validation and scripts/thumbnail_learning.py for managing performance data. These scripts are invoked via the shell to perform deterministic checks and data management tasks.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including video transcripts, user-supplied briefs, and YouTube Studio metrics, to generate hooks and analyze performance. This creates a potential surface for indirect prompt injection if the ingested content contains adversarial instructions.
- Ingestion points: The agent reads user-supplied content (Section 1: transcripts, recordings, notes) and performance metrics (Section 10).
- Boundary markers: While the skill uses structured rubrics and templates, it lacks explicit safety boundaries or delimiters to prevent the agent from following instructions embedded within the source transcripts.
- Capability inventory: The agent has the ability to execute local scripts, write to local data ledgers (performance-ledger.jsonl), and interact with image generation tools.
- Sanitization: The skill documentation mentions using sanitized readbacks for performance data but does not specify validation or filtering steps for the text content of ingested transcripts.
Audit Metadata