The Agent Skills Directory

[SAFE]: The skill implements a secure workflow for local software development, requiring verification of an orchestrator dispatch tag (<ORCHESTRATOR_DISPATCH>) before performing actions. This prevents unauthorized or accidental activation.
[SAFE]: It utilizes standard file system tools (Read, Write, Bash) to manage project artifacts within defined directories (e.g., exploration/prototype/). The logic includes specific detection for "Agent Plugin Mode" to enforce the use of standardized scaffolding tools.
[SAFE]: Human-in-the-loop security is maintained through a mandatory confirmation step. The agent stops and waits for a specific user phrase before proceeding to the handoff phase, ensuring that the generated output is reviewed by a human.
[SAFE]: Reference files and development scripts (such as the Python snippet for keyword analysis) are transparently documented and scoped to local optimization tasks. No suspicious network activity or credential harvesting patterns were found.

subagent-driven-prototyping