The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing Python snippets via python3 -c using direct string interpolation for data variables. In SKILL.md (Phase 4), x-check.md (Phase 3), and x-retro.md (Phase 5), placeholders like POST_CONTENT_HERE and reply_samples=['sample1', ...] are intended to be replaced with live data. If this data contains single quotes or other shell-sensitive characters, it can break the Python string literal and allow for arbitrary command execution within the shell context.
[PROMPT_INJECTION]: The skill processes untrusted data from X.com, specifically post replies and engagement metrics, during the 'Engagement Check' and 'Retrospective' phases. This creates an indirect prompt injection surface where malicious content embedded in a reply could attempt to influence the agent's analysis or strategy updates in Phase 2 of x-retro.md.
[DATA_EXFILTRATION]: The agent uses browser automation to access private account data, including engagement metrics and user replies. While the current implementation saves this to a local SQLite database, the ability to read authenticated browser content represents a significant data exposure risk if the agent's instructions are subverted.
[EXTERNAL_DOWNLOADS]: The skill interacts with x.com and requires the Playwright automation framework. These external interactions are central to the skill's purpose but involve processing content from a remote, third-party service.

x-post