The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches public data from well-known social media platforms and services including Reddit, Hacker News (via Algolia), Bluesky, and YouTube using their official public JSON and RSS endpoints. These operations are conducted for the purpose of social listening and do not involve executable code.
[COMMAND_EXECUTION]: Utilizes standard data processing utilities like jq and xmllint to parse and structure the retrieved social media content. The commands are scoped to data transformation and do not include dangerous flags or shell piping to interpreters.
[DATA_EXFILTRATION]: No evidence of unauthorized access to sensitive local files (such as SSH keys, AWS credentials, or .env files) or exfiltration of private environment variables was found.
[PROMPT_INJECTION]: The skill does not contain instructions that attempt to override the agent's system prompt, bypass safety filters, or disclose internal instructions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external social media feeds to draft engagement comments. While this presents an attack surface where a malicious social post could influence the agent's output, this behavior is inherent to the stated purpose of a social listening tool and is handled within expected operational boundaries.

social