The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from X (Twitter) search results and historical posts to inform its drafting and remixing logic. This creates a surface for indirect prompt injection if an ingested tweet contains malicious instructions aimed at the agent.
Ingestion points: Fetches real-time search results and account history via the x_search tool from api.x.ai.
Boundary markers: The skill does not implement specific delimiters or 'ignore instructions' markers when interpolating external tweet content into its generation prompts.
Capability inventory: The skill possesses capabilities for subprocess execution (curl, jq, date), filesystem persistence (memory/drafts/), and operator notification (./notify).
Sanitization: While the skill employs a 'Quality Gate' for remixes to ensure novelty and voice consistency, it lacks explicit filtering to detect or strip adversarial commands within the ingested text.
[COMMAND_EXECUTION]: The skill utilizes shell commands to manage its logic and data processing.
It uses curl (via the platform's secretcurl) to interact with external APIs and jq for robust JSON manipulation.
It employs standard utilities like date, sed, grep, and cut for managing time windows and parsing arguments. These operations are within the expected scope for a skill of this complexity.
[EXTERNAL_DOWNLOADS]: The skill fetches data and performs searches via XAI's (Grok) official API endpoints. This is a neutral operation targeting a well-known service provider to fulfill the skill's primary purpose.

write-tweet