The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/sync_flow.py executes a local system command via subprocess.run to call gh auth token for optional GitHub authentication. This invocation is securely structured as a hardcoded sequence of arguments with no shell evaluation or user-controlled variable interpolation, preventing command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection since it is designed to ingest and evaluate text or data from external URLs or topics specified by the user.
Ingestion points: Content from user-provided topics or retrieved target URLs processed across commands like /blog flow find and /blog flow optimize.
Boundary markers: The prompts utilize simple brackets such as [ARTICLE, HUB, OR SITE] to position inputs but do not implement strict semantic boundaries or instructions to ignore nested commands.
Capability inventory: The overall workspace possesses file writing and network retrieval capabilities restricted to the repository's synchronization logic.
Sanitization: There is no explicit sanitization or filtering applied to incoming text inputs before being passed into the language model context.

blog-flow