The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions require the agent to use dangerouslyDisableSandbox: true at the start of every invocation to initialize telemetry, which bypasses default security restrictions for shell command execution.
[DATA_EXFILTRATION]: The skill transmits telemetry data to https://events.eu.confidence.dev. This data includes user session identifiers, step timing, and a manual assessment of the user's "sentiment" based on their interaction style.
[REMOTE_CODE_EXECUTION]: During telemetry setup, the skill fetches a client secret from a remote endpoint (https://onboarding.confidence.dev) and pipes it directly into a Python interpreter (python3 -c) for execution.
[DYNAMIC_EXECUTION]: The skill generates a plan file containing executable shell and MCP commands during the planning phase, which are then processed and executed by the /migrate-optimizely execute command.
[COMMAND_EXECUTION]: The skill performs automated execution of shell and MCP commands stored in a local markdown plan file during the migration process.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external APIs (Optimizely) and local source code without explicit sanitization or delimitation instructions.
Ingestion points: Flag definitions from Optimizely API/files and source code files scanned during Phase 2.
Boundary markers: Absent; no instructions for using delimiters or "ignore" warnings.
Capability inventory: The skill can execute shell commands and MCP tools (createFlag, addTargetingRule, resolveFlag) during the execution phase.
Sanitization: Absent; no logic provided to escape or validate external content before inclusion in the plan.

migrate-optimizely