The Agent Skills Directory

[SAFE]: The skill incorporates a comprehensive 'Run Contract' and 'Source Policy' that explicitly requires the redaction of secrets from all artifacts and treats external content as evidence rather than instruction authority.
[SAFE]: The dashboard renderer script (scripts/render-dashboard.mjs) uses a secure temporary file writing pattern and performs validation on the experiment data before processing.
[SAFE]: The dashboard template (assets/dashboard-template.html) includes a custom Markdown parser that enforces a safe subset of tags and mandatory HTML escaping for all dynamic content to prevent cross-site scripting (XSS) vulnerabilities.
[SAFE]: The skill enforces a strict 'one-mutation-at-a-time' protocol with mandatory 'Guard' checks that cannot be bypassed by high evaluation scores, ensuring that safety boundaries are preserved throughout the optimization process.

autoresearch-skill