Subagent Type Catalogue

The 34 role-based subagent types dispatchable via the opencode Task tool. Each type has one tracking issue in this repo, tagged agent-catalogue.

Distinct from the persona agents in Agent-Registry (Kimiko/Kairo/ Shimaguru). Personas are persistent identities; subagent types are capability profiles spun up on demand.

Quick reference

Type	Role	Category
`general`	General-purpose research and multi-step execution	General
`explore`	Fast codebase search (files, keywords, Q&A)	Exploration
`developer`	Implements specs with tests	Engineering
`architect`	Code analysis, solution design, ADRs	Engineering
`backend-architect`	APIs, server logic, databases, scalable backends	Engineering
`frontend-developer`	UI, React/Vue/Angular, state management	Engineering
`cross-platform-builder`	Tauri 2.0, Rust/WASM, desktop + mobile	Engineering
`atomic-data-engineer`	Atomic Data Server, ontologies, @tomic/cli	Engineering
`ai-engineer`	AI/ML features, LLM integration, recommenders	Engineering
`rapid-prototyper`	MVPs, scaffolding, trending features	Engineering
`rust-code-reviewer`	Expert Rust code review	Quality
`rust-performance-expert`	Rust perf, profiling, SIMD, zero-copy	Quality
`quality-reviewer`	Reviews code for real issues (security/data-loss/perf)	Quality
`overseer`	System quality, security, architectural verification	Quality
`development-observer`	Verifies other agents fulfil requirements	Quality
`via-negativa-analyst`	Critical analysis: gaps, risks, failures	Quality
`perspective-investigator`	Multi-faceted analysis before action	Quality
`insight-synthesizer`	Consolidates findings into unified strategy	Quality
`debugger`	Debugs Rust/WASM/streaming pipelines	Quality
`test-writer-fixer`	Writes tests, analyses failures, fixes them	Quality
`technical-writer`	Documentation	Support
`technical-cto`	Technical thought-leadership content	Support
`strategy-executor`	Turns strategies into actionable tasks	Support
`caddy-server-manager`	Caddy web server management	Ops
`devops-automator`	CI/CD, cloud infra, monitoring, deployment	Ops
`phase1-researcher`	Phase 1: research and problem understanding	V-model (left)
`phase2-designer`	Phase 2: planning and design	V-model (left)
`disciplined-specification`	Phase 2.5: specification interviews	V-model (left)
`disciplined-quality-evaluation`	Phase 1.5/2.5: KLS quality gates	V-model (gate)
`left-side-of-v`	Orchestrates Research + Design phases	V-model (left)
`phase3-implementer`	Phase 3: step-by-step implementation	V-model (right)
`disciplined-implementation`	Phase 3: approved plans to tested code	V-model (right)
`execution-orchestrator`	Orchestrates execution with quality gates	V-model (right)
`right-side-of-v`	Orchestrates Verification + Validation	V-model (right)

Dispatch guidance

Single-file edit / quick lookup -> do NOT use a subagent; call the tool directly.
Whole-codebase / >5 interdependent steps -> general or explore.
Planning a new feature -> left-side-of-v (or phase1-researcher -> phase2-designer).
Implementation -> phase3-implementer / disciplined-implementation / developer.
Pre-merge assurance -> right-side-of-v / overseer / quality-reviewer.
Rust-specific -> rust-code-reviewer / rust-performance-expert / debugger.
Adversarial / risk -> via-negativa-analyst / perspective-investigator.

Tracking issues

Each type has an open issue titled Agent: <Name> — <role>, labelled agent-catalogue + type:task + workflow:ready. Filter:

gtr list-issues --owner terraphim --repo agent-tasks --labels agent-catalogue

Created 2026-06-28. See Agent-Registry for persona agents.