The Agent Skills Directory

[SAFE]: The skill enforces a strict 'Artifact Contract' that prohibits external fonts, styles, scripts, or network calls, ensuring that documentation remains self-contained and private.
[DATA_EXPOSURE]: The file operations are restricted to the project's documentation directory (docs/artifacts/). There is no evidence of the skill attempting to access sensitive system files, environment variables, or credentials.
[REMOTE_CODE_EXECUTION]: The provided Python scripts rely exclusively on standard libraries (argparse, html, re, pathlib). No external packages are installed, and no remote code is downloaded or executed.
[COMMAND_EXECUTION]: While the skill involves running Python scripts to create and validate files, these scripts use safe path handling and do not invoke shell interpreters for arbitrary command execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes project data to generate documentation, creating a surface for potential injection. However, the check_artifacts.py script mitigates this by scanning for unauthorized network APIs (e.g., fetch, XMLHttpRequest) and external resource links, providing a secondary defense against data exfiltration attempts hidden in generated HTML.
Ingestion points: Project source files used to derive visualizations (SKILL.md).
Boundary markers: Not explicitly defined in the prompts.
Capability inventory: File write operations via scripts/create_artifact.py and directory creation.
Sanitization: Uses html.escape for metadata and regex-based content validation for scripts and CSS.

harness-artifact