design-system-ops-internal

Produce a six-section, AI/Figma-MCP-optimised spec for one component.

Detect system-wide visual/behavioural/API/token drift and classify each by cause.

Run the full diagnostic agent — deep audit chain + cross-skill synthesis (proposal only).

Run the governance-review agent — internal assessment + leadership brief in one cycle.

Run the migration agent — understand → transformation table → codemod → four-phase rollout.

Audit ONE component for accessibility across keyboard, focus management, screen-reader/ARIA, colour contrast, and motion, mapped to WCAG 2.1 AA, and report audit thoroughness AND accessibility as two separate labels. Triggers on 'is this component accessible', 'a11y check on X', 'audit accessibility of the dialog', 'WCAG check for this component'. Do NOT trigger for theme-wide contrast across all tokens — use theme-audit. Do NOT trigger for a multi-check release gate — use the component-to-release agent. Do NOT trigger for the whole library's coverage — use component-audit.

Report design-system adoption — coverage (of surfaces that could use the system, how many do) reported SEPARATELY from adoption (the usage trend) — with at-risk areas flagged and a baseline established on first run. Triggers on 'adoption report', 'how widely is the design system used', 'which areas are lagging on the design system', 'coverage report'. Do NOT trigger for component library health/quality — use component-audit. Do NOT trigger for a leadership business brief — use stakeholder-brief.

Produce a six-section, LLM- and Figma-MCP-optimised specification for ONE component — identity, API, states/variants, composition, tokens/theming, usage/a11y — so agents and tools can consume it reliably. Triggers on 'document this component for AI', 'write a component spec for X', 'describe the Dialog for the MCP', 'generate component metadata doc'. Do NOT trigger for a composed multi-component pattern — use pattern-documentation. Do NOT trigger for when-to-use guidance only — use usage-guidelines. Do NOT trigger for a machine-readable index of ALL components — use codebase-index.

Convert audit/validate findings into structured, estimated work items ready for a tracker — title, context, acceptance criteria, effort size, and a priority derived from each finding's severity. Triggers on 'turn these findings into tickets', 'make a backlog from the audit', 'generate work items', 'backlog from drift-detection'. Do NOT trigger for producing the findings themselves — run the relevant audit/validate skill first. Do NOT trigger for release messaging — use change-communication.

Produce release communication — release notes, a migration guide, and an announcement — scaled to the impact level of the change (a quiet patch note vs a breaking-change campaign). Triggers on 'write release notes for X', 'announce this change', 'migration guide for the token rename', 'comms for the deprecation'. Do NOT trigger for choosing the version number — use version-bump-advisor. Do NOT trigger for the deprecation plan itself — use deprecation-process (this writes its comms).