The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (tweets) to generate draft responses. This creates a potential vector where malicious instructions embedded in a tweet could influence the agent's drafting logic.
Ingestion points: External tweet content fetched via the X.AI API search tool (SKILL.md).
Boundary markers: No explicit boundary markers or delimiters are used to isolate untrusted tweet content from the agent's instructions during the response generation phase.
Capability inventory: The skill can execute shell commands (curl, jq, date), perform network operations, and read/write to the memory/ and soul/ directories (SKILL.md).
Sanitization: The skill does not specify any sanitization or filtering of the fetched tweet text before it is used as context for the LLM.
[COMMAND_EXECUTION]: The skill uses local shell utilities to manage its workflow and process data.
It utilizes curl (via a ./secretcurl wrapper) to interact with API endpoints.
jq is used to build search payloads and parse JSON responses from the X.AI API.
The skill uses a local ./notify script to deliver drafts to the operator.
[EXTERNAL_DOWNLOADS]: The skill fetches content from a well-known technology service.
It connects to https://api.x.ai/v1/responses to search for and retrieve recent tweets for analysis.

reply-maker