The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The README provides instructions for users to install the skill using the npx skills add Yeadon8888/cangjie-skill command. This involves downloading and executing a package from a remote repository, which is the standard installation method for the platform.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion engine that processes untrusted external data, including user-uploaded files (PDF, TXT, MD, EPUB) and WebSearch results, to generate personality frameworks. This creates a surface for indirect prompt injection if the source materials contain adversarial instructions.
Ingestion points: Documents uploaded by the user and web content retrieved during the research phase.
Boundary markers: The skill manifest (SKILL.md) and the provided references/skill-template.md use structured markdown headers to delimit cognitive context sections.
Capability inventory: The skill utilizes WebSearch, file reading, directory creation, and file writing to generate new persona-based skills.
Sanitization: The distillation instructions do not explicitly require the agent to sanitize or escape extracted content before writing it into the generated skill manifest.
[COMMAND_EXECUTION]: The scripts/ directory contains Python tools (compile-prompt.py and quality_check.py) intended for prompt management and validation. These scripts perform local file system operations, specifically reading skill manifests provided via command-line arguments.

cangjie