The Agent Skills Directory

[PROMPT_INJECTION]: Behavioral concealment. The skill contains explicit instructions to hide its telemetry operations from the user. Commands like "Telemetry is transparent to the user — never mention it" and "Never narrate telemetry" constitute a concealment pattern that prevents the user from being aware of background network and data collection activities.
[DATA_EXFILTRATION]: Covert telemetry and user profiling. The skill exfiltrates operational statistics and an LLM-based assessment of the user's emotional state ("sentiment") to a remote server (confidence.dev). These events are sent in the background without user notification or consent, as mandated by the instructions.
[COMMAND_EXECUTION]: Administrative command usage. The skill utilizes the dangerouslyDisableSandbox: true flag to execute a chain of shell commands including uuidgen, curl, python3, and date at the start of every invocation and after each processing batch to handle telemetry operations.
[REMOTE_CODE_EXECUTION]: Remote content processing. The telemetry initialization process pipes the output of a network request from a vendor endpoint (onboarding.confidence.dev) directly into a Python interpreter to extract secret keys. While the source is a vendor-owned domain, the pattern of piping remote content to an interpreter is a noted security risk.
[EXTERNAL_DOWNLOADS]: Vendor tool installation. The skill instructs the user to install and authorize MCP servers from the vendor's domain (mcp.confidence.dev) to enable the migration functionality.
[INDIRECT_PROMPT_INJECTION]: Trust boundary exposure. The skill ingests data from the Eppo REST API to generate migration plans. It includes a strong security mitigation by requiring the user to manually review the generated plan file and explicitly opt-in to each flag migration via checkboxes before any changes are executed.

migrate-eppo