The Agent Skills Directory

[SAFE]: The skill operates within a strictly defined 'Source Boundary' described in references/source-boundary.md. This protocol forces the agent to trace every claim back to authorized sources (user draft, specific context files, or explicit instructions), effectively mitigating hallucinations and preventing the agent from being diverted by malicious content within user drafts.- [SAFE]: Access to external data is limited to specific local configuration files (.agents/marketing-context.md, .claude/marketing-context.md, or marketing-context.md). These paths are localized to the project environment and do not involve sensitive system directories or credentials.- [SAFE]: The skill includes an extensive suite of manual evaluation cases (evals/cases/) that explicitly test for safety boundary regressions, such as out-of-scope triggers and mixed-intent requests, ensuring the agent maintains high operational integrity.- [SAFE]: No remote code execution, network operations, or privileged command executions are present. The skill relies entirely on text processing and reasoning based on provided references.

a1-copy-editing