The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file utilizes dynamic context injection (the ! syntax) to execute environmental discovery commands at load time. These commands probe for Git status, repository manifests (package.json, docker-compose.yml), and configuration directories (Kubernetes, Terraform). These operations are used solely to populate the agent's context with relevant system topology data for diagram generation and are considered safe.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted data from repository configuration files, such as service names, dependencies, and resource identifiers. A malicious actor with write access to these files could attempt to embed instructions within metadata fields. However, the skill includes a validation script (scripts/validate_html.py) that checks the generated output for unauthorized script injections, and the instructions focus on structured data processing, which limits the risk.
[EXTERNAL_DOWNLOADS]: The skill references external assets from well-known and trusted services. Specifically, the templates/mermaid-fallback.html template loads the Mermaid.js library from cdn.jsdelivr.net, and various templates load typography from fonts.googleapis.com. These are documented and standard practices for the intended functionality.
[COMMAND_EXECUTION]: The skill uses several utility scripts (scripts/collect_dci.sh, scripts/fingerprint.py, scripts/validate_html.py, and scripts/open_in_browser.sh) to automate the workflow of harvesting data, calculating structural hashes, validating HTML, and opening the resulting diagram. These scripts are functional in nature, scoped to the local repository, and do not perform unauthorized network operations or privilege escalation.

engineer-design-diagram