The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of user-supplied topics, angles, and rough drafts. This represents a surface for indirect prompt injection where instructions could be embedded in the input data. However, the skill mandates a human review step (Step 7: Approval card) before any external actions are taken. The available capabilities are limited to publishing via lib.publish and generating images via lib.illustrate.
Ingestion points: User-provided topic, angle, and rough draft; voice-profile.md reference file.
Boundary markers: No specific delimiters are defined for user input interpolation.
Capability inventory: lib.publish (sends text to social media), lib.illustrate (requests image generation).
Sanitization: The skill requires an explicit 'Approval card' shown to the user, and actions are only performed 'On approval'. It also runs a 'humanizer' audit pass to scrub specific linguistic markers.
[DATA_EXFILTRATION]: The skill uses lib.publish to send content to https://x.com/compose/tweet. This network operation targets a well-known service and is the primary, intended purpose of the skill. The target URL is hardcoded in the instructions.

x-post-writer