The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to analyze external codebase manifests and structures, creating a surface for indirect prompt injection.
Ingestion points: The 'Brownfield Mode' described in references/analysis.md reads file structures, package.json, and README.md to assess project complexity.
Boundary markers: SKILL.md defines explicit CANNOT and APPROVAL REQUIRED boundaries to restrict agent actions and prohibit writing business logic.
Capability inventory: Across all reference files, the skill's capabilities are limited to generating markdown-based agent primitives and structural recommendations.
Sanitization: There is no documented mechanism for sanitizing or escaping content ingested from the analyzed files.
[COMMAND_EXECUTION]: Architecture patterns in references/patterns.md provide templates for agents that utilize the terminal and write tools. While the skill itself is instructional and focuses on documentation, it promotes the creation of autonomous agents with high-privilege access to the host system.

prose-architect