The Agent Skills Directory

[SAFE]: The skill demonstrates a strong security posture for a productivity tool.\n
File access in word/profile.js and word/visuals.js is strictly limited to local paths, preventing SSRF or remote data exfiltration.\n
SVG processing in word/visuals.js includes a validator that blocks scripts, iframes, and external resource references to ensure static content rendering.\n
The documentation provides clear guidance on data de-identification and privacy when handling sensitive legal data.\n
Use of disable-model-invocation: true in the frontmatter serves as a safety control to prevent accidental generation of legal documents in general chat sessions.\n- [COMMAND_EXECUTION]: Benign command execution is used for document orchestration and testing.\n
The rendering engine and visual conversion scripts use subprocesses to call standard utilities like the Mermaid CLI (mmdc) and internal Node.js helper scripts.\n
Regression tests in word/tests/test_layout_regression.py utilize subprocess.run to execute Node.js and npm to verify the structure of generated documents.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied legal materials and records, presenting a surface for indirect prompt injection.\n
Ingestion points: Input Markdown or JSON files read by word/render.js.\n
Boundary markers: The skill relies on Markdown headers and structured JSON blocks to delimit user content.\n
Capability inventory: The skill possesses file read/write access and can execute shell commands via Node.js and Python for the document rendering process.\n
Sanitization: Input is validated in word/render.js and word/profile.js to strip HTML tags and reject unexpected code blocks, mitigating the risk of malicious instructions influencing the rendering output.

legal-proposal-generator