claude-code-stack

Use for UI changes — new pages, new components, dashboard work. Audits accessibility — contrast, keyboard navigation, screen reader behavior, focus management, semantic structure, ARIA where needed. Pairs with designer. Important for enterprise (customers ask) and ethical (real users need it).

Designs before code. Produces ADR + interface contracts + test plan. Refuses to write code. Invoked at the start of any non-trivial task (new feature, refactor, schema change, system integration).

Use AFTER architect on novel features, or for any architectural decision that locks the system into a direction. The systemic critique runs through Gemini (via the local Gemini CLI) — a different model family from the architect (Claude), with large context to read the whole repo at once. This subagent orchestrates the Gemini critique and relays it. Skip on routine work. See ADR-012.

Use for schema design, migrations, idempotency design, data quality, ETL/ELT pipelines, normalization/denormalization decisions. Invoked for any change in schema-migration domain mode. Distinct from architect — data-engineer reasons specifically about data shape, constraints, history, and recovery.

Visual + system design. Owns the design system. Default model is Opus because design judgment requires synthesis across visual, functional, and emotional dimensions. Used heavily during dashboard work — the maintainer is rebuilding the Spend Dashboard and turning it into a system applied across all dashboards.

Create or update an Architecture Decision Record. Captures decisions worth more than 5 minutes of thinking, in a format successors can read cold. Use when architect (or user) makes a non-trivial design decision. Output goes to docs/ADRs/<NNN>-<slug>.md.

Run monthly to evaluate how each subagent has actually performed over the last 30 days. Reads subagent_runs table, identifies patterns of failure/success, proposes prompt refinements or model changes. Distinct from /model-audit (which looks outside at benchmarks) — this looks inside at actual outcomes. Targets the assumption that current configuration is still optimal.

Shortcut to flip orchestration mode. /agent-teams on enables agent-teams mode (sets CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in .claude/settings.json and orchestration_mode=agent-teams in stack-config.json). /agent-teams off reverts. /agent-teams hybrid enables hybrid mode. /agent-teams (no arg) or /agent-teams status shows current state.

Per-repo opt-in for GitHub auto-merge gated on a required `run-tests` status check. /auto-merge on <repo> ensures a run-tests CI check exists (scaffolds one if missing), enables allow_auto_merge, creates a per-repo ruleset requiring run-tests, and installs an auto-enable workflow so PRs merge themselves once tests pass. /auto-merge off <repo> reverts. /auto-merge status <repo> shows current state. Use when a repo's PRs should merge automatically on green CI instead of waiting for a manual merge.

Analyze a codebase and recommend Claude Code automations (hooks, subagents, skills, plugins, MCP servers). Use when user asks for automation recommendations, wants to optimize their Claude Code setup, mentions improving Claude Code workflows, asks how to first set up Claude Code for a project, or wants to know what Claude Code features they should use.

11 hooks across 4 events