The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions explicitly demand the removal of human oversight and user confirmation, effectively overriding the agent's default safety preference for human-in-the-loop operation.
Evidence: "Zero Confirmation: Run the entire pipeline without asking for user approval between steps", "You never ask for permission between steps", "Notify Only at End: The user sees results (published URLs), not process updates."
[COMMAND_EXECUTION]: The skill establishes persistence by instructing the agent to modify system scheduling tasks to ensure recurring execution without user intervention.
Evidence: "Self-Schedule: Read learnings.json bestTimes and schedule next execution at the optimal posting time", "adjusts cron schedule so next execution happens at peak engagement hour."
[REMOTE_CODE_EXECUTION]: The skill relies on the execution of a suite of external scripts across multiple environments (Shell, Python, and Node.js) that perform network and file operations.
Evidence: The workflow triggers generate-slides.sh, generate_image.py, publish-carousel.sh, check-analytics.sh, analyze-web.js, and learn-from-analytics.js.
[DATA_EXFILTRATION]: The skill exhibits an indirect prompt injection surface due to the ingestion and processing of untrusted data from third-party websites.
Ingestion points: Playwright-based scraping of target URLs and internal pages (pricing, features, testimonials) within analyze-web.js.
Boundary markers: Absent; the skill does not specify delimiters or warnings to ignore instructions embedded in the scraped content.
Capability inventory: Extensive capabilities including shell script execution, Python and JavaScript execution, and multi-platform network API publishing.
Sanitization: No sanitization or validation logic is mentioned for the content extracted from external URLs before it influences generative prompts or automated actions.

Carousel Growth Engine