Manage agent roster — list agents, show details, check model assignments, verify setup, and dispatch work to the right agent. Covers all 29 agents, 11 teams, and 8 chains.
Source of truth: /mnt/c/Users/shuff57/Documents/GitHub/agent-evo/roster/ Synced to ~/.claude/agents/ and ~/.config/opencode/superpowers/agents/ via sync.sh. (Hermes does not use a dedicated agents dir — agents are invoked via Claude Code or OpenCode.)
| Task | Agent | Model |
|---|---|---|
| General coding | code-engineer | github-copilot/claude-sonnet-4.6 |
| Autonomous implementation | prometheus | github-copilot/claude-sonnet-4.6 |
| End-to-end orchestration | atlas | github-copilot/claude-sonnet-4.6 |
| Strategic planning | planner | github-copilot/claude-opus-4.6 |
| Architecture / hard debug | oracle | github-copilot/claude-opus-4.6 |
| Pre-planning analysis | metis | github-copilot/claude-opus-4.6 |
| Evolution / self-improve | evolver | github-copilot/claude-sonnet-4.6 |
| Quality verification | critic | ollama/glm-5.1:cloud |
| Security / adversarial | red-team | ollama/glm-5:cloud |
| Test suite authoring | qa-tester | ollama/qwen3-coder-next:cloud |
| Codebase exploration | scout | ollama/minimax-m2.7:cloud |
| External docs / libs | librarian | ollama/minimax-m2.7:cloud |
| Meta-coordination | meta-orchestrator | ollama/minimax-m2.7:cloud |
| Debugging / diagnosis | debugger | ollama/kimi-k2.5:cloud |
| UI/UX design | designer | ollama/kimi-k2.5:cloud |
| Screenshots / PDFs / images | visual-analyzer | github-copilot/gemini-3.1-pro-preview |
| Browser automation | bowser | ollama/gemma4:31b-cloud |
| Documentation | documenter | ollama/gemma4:31b-cloud |
| Text summarization | summarizer | ollama/gemma4:31b-cloud |
| Agent loading check | test-ping | ollama/gemma4:31b-cloud |
Read-only research agents. Queried in parallel by meta-orchestrator.
agents-expert — agent .md frontmatter format, teams.yaml, agent-chain orchestration cli-expert — CLI flags, env vars, subcommands, non-interactive usage config-expert — settings files, provider config, model selection extensions-expert — custom tools, event handlers, commands, shortcuts, state management keybindings-expert — shortcut registration, key IDs, modifier combos, terminal compat prompts-expert — prompt template .md format, positional args, discovery locations skills-expert — SKILL.md format, frontmatter fields, directory structure theme-expert — color token systems, theme JSON, variable systems, hot reload ui-expert — built-in/custom components, overlays, keyboard input, widgets
| Category | Agents | Access |
|---|---|---|
| Exploration/Research | scout, librarian | Read-only |
| Planning/Analysis | planner, metis | Read-only |
| Review/Critique | critic, red-team, qa-tester | Read-only |
| Advisory/Debugging | oracle, debugger | Read-only |
| Implementation | code-engineer, prometheus, designer, documenter, summarizer | Read-write |
| Orchestration | atlas, meta-orchestrator, evolver | Read-write |
| Visual/Browser | visual-analyzer, bowser | Read-only |
| Domain Experts | 9 experts (see above) | Read-only |
| Utility | test-ping | Read-only |
| Team | Agents | Use Case |
|---|---|---|
| full | scout, planner, code-engineer, critic, documenter, red-team | Full lifecycle |
| info | scout, documenter, critic | Information gathering |
| frontend | planner, code-engineer, bowser | Frontend work |
| experts | meta-orchestrator + all 9 domain experts | Platform development |
| omo | oracle, librarian, scout, critic | Deep research |
| advisory | oracle, metis, critic | Architecture decisions |
| implementation | code-engineer, prometheus, atlas | Building |
| planning | scout, planner, critic | Plan creation |
| quality | qa-tester, critic, red-team | Post-implementation |
| design | designer, code-engineer, critic | Design work |
| debug | debugger, scout, oracle | Troubleshooting |
| Chain | Steps | Use Case |
|---|---|---|
| plan-build | planner → code-engineer | Fast two-step |
| scout-flow | scout → scout → scout | Triple-scout deep recon |
| plan-review-plan | planner → critic → planner | Iterative planning |
| full-review | scout → planner → code-engineer → critic | Standard lifecycle |
| research-plan-build | librarian → planner → prometheus → critic | Library-dependent tasks |
| oracle-then-build | oracle → prometheus → critic | Architecture-guided build |
| ultrawork | scout → planner → critic → prometheus → critic | 5-phase autonomous loop |
| deep-interview-then-build | metis → planner → critic → prometheus → critic | Vague/complex requests |
| Model | Agents |
|---|---|
| github-copilot/claude-sonnet-4.6 | prometheus, code-engineer, atlas, evolver |
| github-copilot/claude-opus-4.6 | planner, oracle, metis |
| github-copilot/gemini-3.1-pro-preview | visual-analyzer |
| ollama/glm-5.1:cloud | critic |
| ollama/glm-5:cloud | red-team |
| ollama/minimax-m2.7:cloud | meta-orchestrator, scout, librarian |
| ollama/kimi-k2.5:cloud | debugger, designer |
| ollama/qwen3-coder-next:cloud | qa-tester |
| ollama/gemma4:31b-cloud | documenter, summarizer, bowser, test-ping, all 9 experts |
cd /mnt/c/Users/shuff57/Documents/GitHub/agent-evo
bash sync.sh
Symlinks roster/ -> ~/.claude/agents/ and ~/.config/opencode/superpowers/agents/. Run after cloning or when adding/removing agents.
Verify symlinks: ls -la ~/.claude/agents # should point to roster/ ls -la ~/.config/opencode/superpowers/agents # should point to roster/
Count agents: ls ~/.claude/agents/*.md | wc -l # should be 30 (29 agents + README.md)
Ping test: invoke test-ping agent — expect "pong" response
For full details on using teams and chains, load the dedicated skills: hermes-team — all 11 teams, when to use each, invocation, pitfalls hermes-chain — all 8 chains, when to use each, $INPUT/$ORIGINAL vars, pitfalls hermes-model-switch — switching providers and models persistently or per-session
To switch the default model or provider for Hermes, load the hermes-model-switch skill. Quick reference: hermes model # interactive picker hermes chat -m openrouter/gpt-4o # per-session hermes config set model.provider copilot # persistent hermes config set model.default claude-sonnet-4.6