The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources: external files (blog-strategy output text files containing tables), web search responses (SERP titles, descriptions, autocomplete outputs, 'People Also Ask' contents), and user inputs during keyword universe expansion. It does not enforce explicit boundary delimiters (such as specific XML tags or triple backticks) when passing these untrusted datasets downstream to /blog write or when processing table fields internally. While it acts as an aggregator, an adversary poisoning external SERP search results or the loaded strategy files could potentially inject instructions that skew post generation, alter execution steps, or cause downstream tools to misbehave. However, the system contains zero critical tools, and any executed commands run under headless mode configurations via predefined helper extensions. Severity remains low.
[COMMAND_EXECUTION]: The skill orchestrates and schedules sequential executions of external platform sub-skills like /blog write, /blog analyze, /blog image, and /blog schema via the Task tool mechanism. The parameters for these runs are dynamically computed from files like cluster-plan.json or imported tables.
[EXTERNAL_DOWNLOADS]: Evaluates external components via WebSearch to build a keyword universe from various signals and checks Google organic result top-10 URLs for correlation, which does not involve fetching or evaluating dangerous code payloads.

blog-cluster