The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates a series of shell commands from the 'nuke' ecosystem, such as nuke recon, nuke audit, and nuke exec. It manages stage transitions and provides the user with the exact commands needed to progress through a multi-session workflow.
[EXTERNAL_DOWNLOADS]: Reference documentation (preflight.md) describes a mechanism used by specific ceremonies within the ecosystem to resolve 'environmental blockers' by automatically installing missing dependencies or binaries. This behavior is restricted to enabling the toolset and is explicitly forbidden from modifying project source code, configuration, or manifests.
[PROMPT_INJECTION]: The skill ingests untrusted user input during its interview phase (specifically for project 'scope' and 'goal') and interpolates these values into the .nuke/itinerary.md file as command arguments. Although the skill is instructed to resolve these inputs to concrete paths or git references, this interpolation represents a surface for indirect prompt injection.
Ingestion points: Phase 1 interview questions collecting 'goal' and 'scope' parameters (SKILL.md).
Boundary markers: There are no explicit delimiters or sanitization warnings defined for the command strings generated in the itinerary table.
Capability inventory: The skill possesses the ability to execute various shell commands within the 'nuke' toolset (SKILL.md, Phase 3).
Sanitization: The instructions mandate that 'scope' must be resolved to a concrete path or reference before being added to the itinerary, providing a basic layer of validation against arbitrary command injection.

nuke-pilot