a11y
01Command
Accessibility audit (WCAG 2.2 AA) on the working-tree diff or a specified file/directory. Delegates to @a11y-auditor.
Role-aware Claude Code plugin: 10 specialist agents (frontend, backend, QA, a11y, perf, security, architect, debugger, code-reviewer, docs), 15 domain commands, 8 skills, and a SQLite-backed learning crystal that compounds your corrections across sessions.
Command
Accessibility audit (WCAG 2.2 AA) on the working-tree diff or a specified file/directory. Delegates to @a11y-auditor.
Command
Draft an Architectural Decision Record (ADR) for a proposed or accepted decision. Uses the MADR-style template and places it in docs/adr/ or adrs/ (whichever the repo uses).
Command
Set or inspect token $ budgets. Per-session, daily, and monthly caps. The Stop hook reads these and nudges you when you cross 80% / 100%.
Command
Quality-gated commit — runs lint/typecheck/test (if configured), reviews the staged diff, drafts a conventional-commits message, and commits.
Command
Snapshot the current session's context footprint — transcript size, CLAUDE.md size, MCP schemas, file reads still in context. Identifies what's eating tokens so you can trim.
Command
Session + rolling $ and token usage across all Claude Code sessions — totals, per-model breakdown, cache hit rate, budget status. Data comes from the Stop hook that logs every session.
Command
Multi-phase feature development — Research → Plan → Implement → Review. Delegates research/review to specialist agents; you drive the implementation.
Command
Self-check the holocron plugin install — manifests, SQLite, directories, Node version, hook wiring.
Command
Hunt a flaky test — diagnose non-determinism, race conditions, order dependence, timing, or environment leaks. Delegates to @qa-engineer.
Command
Write a session handoff document summarizing what was done, what's open, and what the next session should pick up. Saves to .holocron/handoff-YYYY-MM-DD.md.
Command
Correction heatmap for the holocron — top tags, weekly additions, oldest rules (review candidates). Makes the crystal visible.
Command
Capture a correction as a durable learning in the holocron (SQLite). Survives compaction, loads on session start, matched against future prompts.
Command
AI-powered verification gate before a destructive or high-stakes operation. Use before git push to main, production deploys, DDL migrations, or mass deletes. Reads the proposed action in context and returns a go / hold decision with reasons.
Command
Audit the MCP servers active in this session — tool count, schema size, and which ones probably aren't pulling their weight. Every MCP tool's schema adds tokens to every turn; disabling the unused ones is a direct cost cut.
Command
Detect the project stack and draft a repo-appropriate CLAUDE.md (language, framework, test runner, lint setup, agents and commands that fit).
Command
Performance audit on the diff or a path. Delegates to @perf-engineer. Flags bundle bloat, N+1 queries, missing indexes, LCP/CLS risks.
Command
Analyze your Claude Code permission setup — flag commands that probably deserve an allowlist entry, dangerous defaults that should be on the denylist, and suggest additions to settings.json. Reduces prompt fatigue without loosening safety.
Command
Enter plan mode with a structured plan scaffold (goal, success criteria, risks, files to touch, rollback). Use before any non-trivial implementation.
Command
Search the holocron (SQLite + FTS5) for relevant past learnings. Returns top matches ranked by bm25.
Command
Staff-level review of the current diff (or a specified ref). Delegates to @code-reviewer, @security-reviewer, and @a11y-auditor as relevant.
Command
Security scan on the diff — SAST heuristics, secret regex sweep, dependency advisory check. Delegates to @security-reviewer.
Command
Find untested branches and edge cases in a file or the diff. Delegates to @qa-engineer. Returns a prioritized list of missing tests with draft names and assertions.
Command
STRIDE-driven threat model of a diff, component, or proposed feature. Delegates to @threat-modeler. Produces an actionable threat list with mitigations — different from /holocron:sec-scan (line-level SAST) in that it works at the design level.
Command
Set up a git worktree for parallel Claude Code sessions — zero-dead-time while one feature bakes. Creates the worktree, creates/switches the branch, and prints the launch command.