Agent
Industry-grade OpenCode all-rounder — your primary agent. 5-gate hybrid workflow with contract-first wave fan-out (Wave 0 contract → Wave 1 parallel builders → Wave 2 parallel verifiers), auto-recovery, pre-warmed dispatch, batched verification. OpenCode-only, never Claude Code.
Agent
Agent/command/skill builder. Reads existing agent patterns, scaffolds new agents, validates agent files against schema. Meta-capability — agent-chadi uses this to build its own team.
Agent
Architecture and integration planning subagent for medium/high-risk changes.
Agent
Backend/API implementation helper for routes, services, validation, auth, and server behavior.
Agent
Dedicated decision council subagent. Four voices (Architect, Skeptic, Pragmatist, Critic) deliberate ambiguous decisions internally. Remembers past decisions via PMB memory for consistency across sessions.
Agent
Database and data-flow helper for schema, migrations, queries, indexes, Prisma/Supabase/Postgres/MySQL/Redis, and data integrity.
Agent
DevOps/release helper for environment variables, deployment config, CI checks, build commands, release notes, and rollback planning.
Agent
Documentation subagent for README, API docs, inline comments, CHANGELOG, migration guides, and project documentation writing.
Agent
Read-only repo explorer for architecture, affected files, dependencies, route maps, and codebase context.
Agent
Frontend/UI implementation and verification helper for layout, forms, routing, responsiveness, and client behavior.
Agent
Wave 2 integration specialist — joins parallel lanes: swaps frontend mocks for real API calls, wires env/credentials, applies local migrations, boots the app and walks the happy path end to end. Resolves contract drift instead of hiding it.
Agent
Memory subagent — graphyloop memory is the active store (PMB MCP disabled). Recall before work, store after.
Agent
Performance and reliability review subagent for render performance, bundle size, API latency, database queries, caching, retries, loading states, and reliability.
Agent
Code quality subagent for linting, formatting, type checking enforcement, and code convention audits. Runs linters, formatters, and type-checkers; fixes violations; enforces project code standards.
Agent
Safe cross-file refactoring agent. Renames, symbol extraction, module restructuring, pattern-based code transformation. Uses ast-grep for structural search/replace and impact analysis. Dry-run first, verify no breakage. Hard refuses new features or behavioral changes — refactor only.
Agent
Final code reviewer for correctness, maintainability, regressions, config validity, security gaps, missing tests, and accidental unrelated changes.
Agent
Security review subagent for auth, RBAC, inputs, uploads, secrets, APIs, database access, XSS, CSRF, CSP, dependencies, and deployment exposure.
Agent
Testing and verification subagent for unit, integration, e2e, build, lint, typecheck, and regression checks. Follows browser-last discipline — vitest+jsdom first, browser tools only for critical E2E.
Agent
Deep thinking/planning subagent. Spawned by agent-chadi when task needs complex reasoning, architecture design, or hard problem-solving. Returns plan — does not code.
Agent
Vision subagent using MoMimo-V2.5 for image processing. Read-only — describes images/screenshots/sketches/diagrams. agent-chadi dispatches when user attaches image or asks about visual content.
Agent
Surgical 1-2 file edit. Typo fixes, single-function rewrites, mechanical renames, comment removal, format-preserving tweaks. Hard refuses 3+ file scope. Returns caveman diff receipt. Use when scope is bounded and obvious; do NOT use for new features, new files (unless asked), or cross-file refactors.
Agent
Minimal fix agent. Reads error output, makes 1-file surgical fix. No discovery, no exploration, no context beyond the error itself. Use when test fails, build breaks, or lint error needs fixing. Hard refuses multi-file changes, new features, or speculative refactors. Faster than graphcrew-builder for pure fix work …
Agent
Read-only code locator. Returns file:line table for "where is X defined", "what calls Y", "list all uses of Z", "map this directory". Output is caveman-compressed so the main thread eats 60% fewer tokens than vanilla Explore. Refuses to suggest fixes.
Agent
Diff/branch/file reviewer. One line per finding, severity-tagged, no praise, no scope creep. Output format path:line: : . . Use for "review this PR", "review my diff", "audit this file". Skips formatting nits unless they change meaning.