The Agent Skills Directory

[COMMAND_EXECUTION]: In lib/backend_selector.py, the skill can execute a user-defined shell command provided via the X_SKILLS_CUSTOM_POSTER environment variable. While this is an intended feature for custom posting backends and uses shlex.split for argument parsing rather than shell=True, it remains a high-privilege capability that executes local binaries.
[DYNAMIC_EXECUTION]: The publish function in lib/backend_selector.py dynamically assembles and runs a command-line string based on environment configuration to delegate post publishing to external scripts.
[INDIRECT_PROMPT_INJECTION]: The skills x-audience-insights, x-hook-extractor, and x-reply-drafter ingest untrusted data from X (tweets and replies) either through direct user paste or the Apify API (lib/apify_client.py).
Ingestion points: Tweet and reply text enters the agent context through ApifyClient methods or manual user input.
Boundary markers: Explicit boundary markers for external data are absent in the script logic, though the system relies on markdown instruction constraints.
Capability inventory: The agent has capabilities for network access, file writing (to references/voice-profile.md), and arbitrary command execution (via the DIY poster).
Sanitization: No explicit sanitization or filtering of ingested tweet content is implemented before processing by the model.
[EXTERNAL_DOWNLOADS]: The skill bundle makes network requests to several external service providers:
api.publora.com for content distribution.
api.apify.com for data extraction.
api.pixfaro.com for image generation. These are documented as primary integrations for the skill's marketing functionality.

x-marketing