The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The /blog flow sync command fetches data from a third-party GitHub repository (github.com/AgriciDaniel/flow) that is not associated with the author's verified infrastructure or the trusted vendors list. These downloaded files are used as executable prompt instructions.
[COMMAND_EXECUTION]: The skill invokes a local script scripts/sync_flow.py via the shell to perform updates. This script performs network operations and writes to the local filesystem, representing a command execution surface for maintenance tasks.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its ingestion of external data.
Ingestion points: The /blog flow optimize [url] command and templates like conversion-audit-prompt.md ingest content from external websites and user-provided document samples (e.g., 'Page or funnel copy').
Boundary markers: The prompt templates provided in the references/ directory do not utilize delimiters or explicit instructions to ignore malicious directives that might be embedded in the analyzed web content.
Capability inventory: The skill possesses capabilities for subprocess execution (python3) and localized filesystem writes.
Sanitization: There is no evidence of input validation or sanitization of the external URL content before it is processed by the agent.

blog-flow