The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-provided prompt goals and existing prompt files to generate hardened instructions. This ingestion surface is managed through a specialized 'breaker wave' phase designed to identify and fix adversarial inputs. * Ingestion points: Arguments <goal> and <path-to-existing-prompt> in Phase 0. * Boundary markers: Structured Phase 1 requirements and Phase 2 mapping are used to separate intent from execution content. * Capability inventory: File system access for writing artifacts to the .nuke/ directory and user-specified destinations; model invocation for specialized 'architect' and 'breaker' roles. * Sanitization: The primary mechanism for safety is Phase 3, which executes a wave of adversarial 'breakers' to test for injections, contradictions, and malformed inputs before the prompt is finalized.
[EXTERNAL_DOWNLOADS]: The shared preflight protocol (in references/preflight.md) provides a mechanism for installing missing environment dependencies if tooling is blocked. This represents standard environment management for the vendor's tool suite.

nuke-prompt