The Agent Skills Directory

[Content Sensitivity and Secret Management]: The skill integrates proactive checks for PII and secrets within the documentation workflow. By providing specific guidance on redaction and the use of placeholders, it helps prevent the accidental exposure of sensitive information in documentation artifacts.
[AI Guardrails and Indirect Injection Safeguards]: Explicit guardrails are provided to ensure the agent treats external documentation as data rather than instructions. This directly addresses the risk of indirect prompt injection. The skill implements a clear evidence chain for these controls: (1) Ingestion points: Processes repository markdown and documentation targets (SKILL.md); (2) Boundary markers: Utilizes injection-boundary and attribution guidance (references/rai-guardrails.md); (3) Capability inventory: Limited to local validation, linting, and formatting commands (references/validation-toolchain.md); (4) Sanitization: Human review is mandatory for changes affecting product claims or policy language.
[Operational Boundaries and Stop Rules]: The skill defines strict 'Stop rules' and 'Handoff triggers' that ensure the agent does not overreach, particularly for tasks involving formal security or accessibility assessments, which are routed to human specialists.
[Local Tooling Scoping]: Commands are restricted to established local repository scripts for documentation linting and testing. This maintains a clear operational scope and avoids the use of untrusted external binaries or remote code execution.

documentation