The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The automated scanner flagged a remote code execution pattern in scripts/tests/test_validate_html.py. However, technical review shows this is a false positive. The file contains a unit test that provides intentionally 'malicious' string samples (e.g., fetch('/api'), XMLHttpRequest) to verify that the skill's validator correctly identifies and blocks them. The subprocess.run call is used solely to execute the local Python validator script for testing purposes.
[COMMAND_EXECUTION]: The skill uses subprocess.run in its testing suite (scripts/tests/test_validate_html.py) and instructs the agent to run scripts/validate_html.py via the CLI. These executions are scoped to the skill's own validation logic and are used to ensure the generated HTML meets security and structural contracts.
[DATA_EXFILTRATION]: The skill explicitly forbids network access. The scripts/validate_html.py script implements regex checks (NETWORK_PATTERNS) to block http, https, @import, fetch, XMLHttpRequest, and WebSocket within the generated output to ensure files remain offline and self-contained.
[INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface as it reads local markdown files to generate visuals. However, the workflow includes a 'Confirmed Input Gate' and a 'Visual Fact Model' step that requires the agent to verify technical facts before generation, reducing the risk of malicious instructions in source files being executed as commands.

technical-visual-companion