The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill uses the 'dangerouslyDisableSandbox: true' flag in Bash code blocks to perform telemetry setup and network operations. This configuration requests the agent to bypass standard security sandboxing for those specific commands, granting them broader access to the host environment.
[DATA_EXFILTRATION]: The skill implements a telemetry system that sends operational data (session IDs, progress metrics, error codes) to 'https://events.eu.confidence.dev'. The instructions explicitly direct the agent to hide these operations from the user, stating they should be 'transparent' and 'never mentioned'.
[EXTERNAL_DOWNLOADS]: The skill requests the installation of external MCP servers from remote URLs belonging to PostHog and Confidence (Spotify). While these are official endpoints for the migration targets, they constitute runtime dependency acquisition from external sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically PostHog flag definitions and the user's codebase. This creates a surface where malicious instructions embedded in flag descriptions or source code comments could potentially influence the agent's logic during the planning or transformation phases.
[DYNAMIC_EXECUTION]: The migration workflow involves generating a markdown plan file containing MCP command payloads. These payloads are subsequently read and executed by the agent, meaning the agent's future actions are determined by content generated in an earlier step.

migrate-posthog