The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill generates reports that reference external CSS and JavaScript libraries from JSDelivr, an established and well-known CDN service. These references are pinned to specific versions (Tailwind v4.3.3 and Mermaid v11.17.1), which minimizes the risk of supply chain attacks.
[COMMAND_EXECUTION]: The skill uses python3 to run a local validation script (scripts/validate_report.py). Additionally, the testing suite (scripts/test_validate_report.py) utilizes subprocess.run to invoke the validator. These executions are scoped to the skill's own logic and testing, representing standard operational behavior for auditing and quality assurance.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the codebase being audited, which presents an indirect prompt injection surface. The skill proactively mitigates this risk by enforcing a strict security contract on its output via the validate_report.py script, which sanitizes the generated HTML against a whitelist of safe tags, attributes, and URLs.
Ingestion points: Source code, configuration files, and repository history of the audited project (processed in SKILL.md).
Boundary markers: The skill uses a post-processing validation step to ensure the output remains within a safe structural and accessibility contract as defined in references/report-contract.md.
Capability inventory: File system access (read/write) and local command execution (python3 for validation).
Sanitization: The scripts/validate_report.py script performs rigorous checks, rejecting active content (such as <iframe> or unauthorized <script> tags), unsafe CSS (such as url() declarations or backslash escapes), and non-whitelisted external network requests.

x9-architecture-scout