The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources including social media platform metadata (YouTube, TikTok, Instagram) and user-uploaded Excel spreadsheets. This represents a potential surface for indirect prompt injection if malicious instructions are embedded in creator profiles or data files.
Ingestion points: Data enters the context via creator search results, profile reads, and spreadsheet imports (monitor import-file, collection import-file, crm import-file).
Boundary markers: The skill includes explicit instructions for the agent to avoid drafting outreach copy independently and to seek user approval before any write operation.
Capability inventory: The skill uses the noxinfluencer CLI which has capabilities for network communication, file exports, and sending emails/messages.
Sanitization: The instructions emphasize human-in-the-loop verification (e.g., dry-runs and explicit approval) rather than automated content filtering.
[CREDENTIALS_SAFE]: The skill provides instructions for secure authentication, specifically advising against exposing API keys in logs or command arguments, and recommending the use of auth --key-stdin or device flow login.
[COMMAND_EXECUTION]: The skill utilizes a dedicated CLI tool (noxinfluencer). All command executions are scoped to this vendor-provided utility, and the instructions require the agent to use --dry-run previews and seek explicit --force approval for state-changing operations.

noxinfluencer