The Agent Skills Directory

[Prompt Injection] (SAFE): The skill instructions focus on documentation structure and professional standards. There are no attempts to bypass safety filters or override system-level constraints.
[Data Exposure & Exfiltration] (SAFE): The skill is designed to read codebase configurations (e.g., application.yml, .env.example) to document architecture. While it interacts with files that might contain secrets, there is no logic for hardcoding credentials or transmitting data to external, untrusted domains. Network usage mentioned in guides is for standard health checks or illustrative examples.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): Although the skill has access to the Bash tool, its workflows are limited to Grep, Glob, and Read operations for analysis. Installation commands (e.g., npm install, git clone) appear strictly within documentation templates and examples intended for user-generated documents, rather than being executed by the skill itself on the host.
[Indirect Prompt Injection] (LOW): The 'brownfield' workflow (code-to-docs) ingests untrusted data by reading source code files to extract API endpoints and data models.
Ingestion points: Files read during codebase analysis (e.g., references/workflows/brownfield-workflow.md).
Boundary markers: None explicitly defined in the prompts to distinguish code content from documentation instructions.
Capability inventory: Read, Write, Edit, Glob, Grep, and Bash tools.
Sanitization: No specific sanitization logic is present for the ingested code content.
[Persistence & Privilege Escalation] (SAFE): No patterns for persistence or unauthorized privilege escalation were identified. Documentation examples mentioning sudo or chmod are contextual templates for users.

documentation-specialist