The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of malicious prompt injection, DAN-style instructions, or safety filter bypass attempts was found. The instructions focus on structured automation tasks.
[DATA_EXFILTRATION]: The skill communicates with twitr.sh to perform social media actions. It includes specific safety guidance to keep user credentials out of the agent context by using external authentication links, which mitigates the risk of credential harvesting.
[REMOTE_CODE_EXECUTION]: The skill mentions adding related tools from the developer's repository (lnvestor/twitr-skills). As these resources belong to the verified author, this is considered a standard operational dependency and does not pose an unexpected remote code risk.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with untrusted external data by monitoring and searching Twitter content, which is a potential surface for indirect injection.
Ingestion points: External data enters the agent context through the x_read, x_search, x_timeline, and x_monitor tools defined in SKILL.md.
Boundary markers: While specific data delimiters are not mentioned, the skill mandates human-in-the-loop confirmation for any resulting actions, providing a significant safety barrier.
Capability inventory: The skill possesses the capability to perform authenticated writes (x_write), including posts and direct messages, based on processed data.
Sanitization: No explicit sanitization or escaping of tweet content is described; however, the requirement for user approval before execution mitigates the risk of automated exploitation.

twitter-automation