Agent Claude Code
Fast mechanical code applier. Applies SEARCH/REPLACE blocks exactly as instructed without reasoning or modification. Use when another agent generates precise SEARCH/REPLACE instructions that need to be applied to files. Never invoke for tasks requiring judgment or code generation.
Agent Claude Code
Context firewall for codebase exploration. Sweeps the project wide (glob/grep across hundreds of candidate files, selective reads) inside its own DISPOSABLE context and returns only the few files/locations that matter, with file:line pointers and minimal excerpts — so the calling agent never burns its expensive…
Agent Claude Code
Judge panel that evaluates Drupal work from seven expert perspectives (correctness, security, Drupal quality, performance, accessibility, translations, test coverage). PLAN mode: invoke BEFORE implementing significant changes — the judges deliberate and return an Implementation Contract (approach, roadmap…
Agent Claude Code
Drupal 10 backend development specialist. Use when creating or modifying custom modules, services, entities, forms, plugins, routing, or hook implementations. Handles PHP backend development only — delegates theming to drupal-theme and visual testing to visual-test. Generates SEARCH/REPLACE blocks and delegates file…
Agent Claude Code
Agent for generating automated tests in Drupal 10/11 projects. Analyzes the code under test, decides the appropriate test type according to the rules defined in drupal-testing, and uses the corresponding skill to generate it. Supports Unit, Kernel, Functional, FunctionalJavascript, Behat and Playwright.
Agent Claude Code
Drupal 10 frontend developer for themes, Twig templates, JavaScript, CSS/SCSS, and TailwindCSS. Use for theming tasks: template creation, preprocess functions, library definitions, asset optimization, responsive design, and Tailwind components. Delegates template quality audits to twig-audit and visual verification to…
Agent Claude Code
Validates code, configurations, or plans generated by other agents. Reviews for correctness, completeness, edge cases, security issues, and Drupal best practices. Use after a development agent produces output that needs quality assurance before presenting to the user. Returns PASS / NEEDS REVISION verdict with…
Agent Claude Code
Ralph Loop requirements planner. Transforms user requests into detailed, structured requirements.md files optimized for autonomous execution with Beads task tracking. Use when the user says "prepare for Ralph", "generate requirements", or wants autonomous overnight execution. Ensures zero ambiguity so Ralph can work…
Agent Claude Code
Delegation advisor for the orchestrator. Decides which specialized agent should handle a task. Use when you are NOT sure which agent fits a request, when a request spans multiple agents, or when a complex task needs to be decomposed before delegating. Do NOT use it when the routing is obvious — delegate directly.…
Agent Claude Code
Interprets the CONTENT of images and screenshots using a vision-capable model: UI/design reviews, layout and spacing critique, accessibility contrast checks, before/after visual comparisons, and extracting information from screenshots (error messages, data in images). Use for design tasks and visual interface…
Agent Claude Code
Captures screenshots and performs visual verification of Drupal pages using Playwright MCP. Use after making frontend changes (templates, CSS, JavaScript) to verify the UI renders correctly. Handles authentication via drush uli for admin pages. Returns screenshots and PASS/FAIL verdicts with visual findings.