The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute shell commands for development workflows, including npm install, npm run lint, npx eslint, npx stylelint, and npm run build as part of its 'Checkfix' cycle.
[COMMAND_EXECUTION]: It supports remote development environments by executing commands via SSH (e.g., ssh nas "cd ..."), which involves handling remote paths and connection parameters.
[EXTERNAL_DOWNLOADS]: The workflow involves running npm install, which downloads third-party dependencies from the npm registry. This is a standard but inherent risk in modern development environments.
[PROMPT_INJECTION]: The skill contains persona-driven instructions (e.g., 'Creative Director', 'Frontend Engineer') and uses strong directives like 'REJECT mechanical dictionaries' and 'OPEN artistic synesthesia' to shape AI behavior, though no malicious bypasses were detected.
[DATA_EXPOSURE]: The skill maintains persistent state and design decisions in a local .debug/ directory. While this is for workflow continuity, it involves frequent reading and writing of project-specific metadata.
[INDIRECT_PROMPT_INJECTION]: As a code-modifying agent, it has a significant attack surface by ingesting untrusted source code and user-provided design descriptions.
Ingestion points: Reads .tsx, .vue, .svelte files, and .debug/ markdown logs.
Boundary markers: No explicit delimiters are defined to separate user-provided data from system instructions during processing.
Capability inventory: Full file-write access and subprocess execution (npm, npx, ssh).
Sanitization: No specific sanitization or validation logic is described for the content it modifies or documents.

debug-ui