The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates subagents and shell commands for complex tasks such as codebase-wide migrations and audits. These capabilities are governed by thorough 'Risk and Cost Gates' that require explicit user approval for budget limits, write scopes, and the initiation of long-running resources like background processes or containers.
[DATA_EXFILTRATION]: No patterns of unauthorized data exfiltration or harvesting were found. The skill's safety policies explicitly prohibit the inclusion of sensitive credentials, secrets, or private customer data in subagent prompts and the durable ledger artifacts.
[PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface because it processes untrusted codebase data and synthesized results from subagents. It mitigates this risk by enforcing a 'Reduce and Cross-check' protocol, requiring that all results be verified against primary evidence (source files, tests, logs) and processed through structured templates to maintain logical boundaries.

ultracode