The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by monitoring public Twitter @mentions, which serves as a significant attack surface.
Ingestion points: External content is retrieved via the twitter_get_mentions tool as described in SKILL.md.
Boundary markers: The skill documentation does not define specific delimiters or instructions to treat mention content as untrusted data during processing.
Capability inventory: The agent has the capability to autonomously post responses to these mentions using the twitter_heartbeat_cycle tool.
Sanitization: A guardrails.py component is mentioned to filter sensitive data like real IP addresses and credentials from outputs, though it may not prevent logic-based injection attacks.
[COMMAND_EXECUTION]: The skill performs autonomous network-based actions (posting tweets) which increases the risk of unintended behavior if the agent is compromised by malicious input.
The twitter_heartbeat_cycle workflow executes automated replies and heartbeat posts without requiring a human-in-the-loop for every individual communication.
The skill uses a safety-first configuration where TWITTER_HEARTBEAT_ENABLED defaults to false, requiring manual operator intervention to activate the autonomous features.

twitter-heartbeat