The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from Vue files located within the project workspace, creating a potential vector for indirect prompt injection.
Ingestion points: scripts/scan_patterns.py and scripts/diff_violations.py use Path.read_text() to ingest the content of .vue files discovered via globbing.
Boundary markers: Extracted data is emitted as structured JSON, but the skill lacks specific boundary markers or instructions to the agent to disregard potential instructions embedded within the extracted patterns (such as within component names or import paths).
Capability inventory: While the tools themselves are read-only, the skill instructions in SKILL.md direct the agent to "Propose small PRs" based on the findings, allowing maliciously crafted patterns in a project file to potentially influence the agent's code generation.
Sanitization: The scripts use regular expressions to filter for specific design tokens and imports, which provides a layer of sanitization, but does not prevent the inclusion of arbitrary strings captured by the regex.
[COMMAND_EXECUTION]: The shared utility script scripts/_skill_cli.py contains functionality for executing system commands.
Evidence: The run_command function utilizes subprocess.run to execute arbitrary commands. Although this function is currently not called by the active scripts (scan_patterns.py or diff_violations.py), its presence in the skill's execution environment provides a capability that exceeds the skill's stated design-auditing purpose.

design-system-generalizer