Agent Reference

Agents are dispatched automatically by Lavra’s workflow commands — you don’t invoke them directly. They run as subagents with specialized instructions for a specific domain.

See the Command Map for a visual overview of how all commands, agents, and skills connect.

Review (16)

Dispatched by /lavra-review, /lavra-work, and /lavra-ship to catch issues before code ships.

architecture-strategist

Evaluates system design, component boundaries, SOLID compliance, and dependency structure.

code-simplicity-reviewer

Identifies unnecessary complexity, premature abstractions, and YAGNI violations.

security-sentinel

Audits input validation, SQL injection, XSS, auth/authz, hardcoded secrets, and OWASP Top 10.

performance-oracle

Finds bottlenecks, N+1 queries, memory leaks, and projects impact at 10x/100x scale.

goal-verifier

Three-level check (Exists, Substantive, Wired) that implementation actually delivers the bead’s success criteria.

kieran-rails-reviewer

Rails code review for conventions, clarity, and maintainability.

kieran-python-reviewer

Python review enforcing type hints, Pythonic patterns, and module organization.

kieran-typescript-reviewer

TypeScript review enforcing no-any policy, type safety, and modern TS 5+ patterns.

dhh-rails-reviewer

Brutally honest Rails review from DHH’s perspective — flags anti-patterns and unnecessary abstractions.

data-integrity-guardian

Reviews migrations, data models, and DB mutations for safety, constraints, and referential integrity.

data-migration-expert

Reviews data backfills and production transformations — validates ID mappings and rollback safety.

deployment-verification-agent

Produces pre/post-deploy checklists, SQL verification queries, and rollback procedures.

migration-drift-detector

Detects schema changes not caused by migrations in the PR (Rails, Alembic, Prisma, Drizzle, Knex).

pattern-recognition-specialist

Finds design patterns, anti-patterns, naming inconsistencies, and architectural boundary violations.

julik-frontend-races-reviewer

Reviews JS and Stimulus for race conditions, timer issues, and Hotwire/Turbo compatibility.

agent-native-reviewer

Checks that user actions have agent equivalents and agents see what users see.

Research (5)

Dispatched by /lavra-design and /lavra-research to gather context before planning.

best-practices-researcher

Researches external best practices, docs, and examples for any technology or framework.

framework-docs-researcher

Fetches framework and library documentation via Context7, checks for API deprecations.

git-history-analyzer

Analyzes git history to trace code origins, identify contributors, and extract development patterns.

learnings-researcher

Searches knowledge.jsonl for past solutions, patterns, and gotchas relevant to the current work.

repo-research-analyst

Analyzes repository structure, architecture files, GitHub issues, and contribution patterns.

Design (3)

Dispatched by /lavra-qa and design-related workflows.

figma-design-sync

Detects and fixes visual differences between web implementation and Figma design.

design-implementation-reviewer

Verifies UI implementations match Figma specs after components are created or modified.

design-iterator

Iteratively refines UI design through screenshot-analyze-improve cycles.

Workflow (5)

Used across various commands for specific task types.

bug-reproduction-validator

Systematically reproduces reported bugs and confirms whether behavior deviates from expected.

spec-flow-analyzer

Maps all possible user flows from a spec, identifies gaps and ambiguities.

pr-comment-resolver

Addresses pull request review comments by implementing the requested changes.

lint

Runs linting and code quality checks on Ruby and ERB files.

every-style-editor

Reviews and edits text content against Every’s house style guide.

Docs (1)

ankane-readme-writer

Creates or updates README files following Ankane-style template for Ruby gems.