Run efficiently
- Provide the raw text or file path plus intended audience and tone goals.
- Ask for specific concerns (clarity, citations, style) if needed.
- Paste the JSON + Markdown output into PRs or review threads for traceability.
Quick reference
Keep this matrix handy to decide which baseline agent to call before escalating to specialised personas.
| Agent | When to use | Ideal outputs |
|---|---|---|
root/brutal-critic.md | Final editorial/code review pass before publishing | JSON scorecard, punchy critique, rewrite ideas |
root/general-research-agent.md | Deep research with citations and structured comparisons | Markdown report, JSON, CSV exports |
root/home-lab-guru.md | Designing or debugging personal lab/edge infrastructure | Topology plan, automation scripts, hardware list |
root/reliability-commander.md | Quality gates, readiness reviews, incident preparation | Checklists, risk matrix, escalation plan |
root/session-closer.md | Closing focused work blocks or sprints | Structured summary, deliverables, next steps |
root/session-restorer.md | Resuming after interruption or onboarding a handoff | Context recap, resume plan, TODO tracker |
brutal-critic.md — Editorial hit squad
Deliver a merciless review of prose, docs, or generated code comments to expose gaps before production.
CLI snippets
claude-code start --file AI-toolkit/agents/root/brutal-critic.md --stdin < doc.mdgemini prompt --file AI-toolkit/agents/root/brutal-critic.md --input article.txtcodex agent AI-toolkit/agents/root/brutal-critic.md --input press_release.mdopencode exec --agent AI-toolkit/agents/root/brutal-critic.md --args "file=summary.md"
Combine with
content-writer.mdfor rewrite passes.session-closer.mdto log critiques in your session history.
general-research-agent.md — Comparison and sourcing engine
Normalise findings from web research, compare competitors, and gather citations ready for briefs or decision docs.
Run efficiently
- State the scope (e.g., “cloud GPU providers for startups”).
- Request output formats such as Markdown, JSON, or CSV.
- Seed with internal links or notes for richer synthesis.
CLI snippets
gemini prompt --file AI-toolkit/agents/root/general-research-agent.md --input "query: top observability tools | format: md,json"claude-code start --file AI-toolkit/agents/root/general-research-agent.md --prompt "Compare ML experiment platforms"codex agent AI-toolkit/agents/root/general-research-agent.md --input research_brief.md
Combine with
create_brief.mdprompt for product briefs.reliability-commander.mdwhen findings impact release criteria.
home-lab-guru.md — Personal infrastructure architect
Design or troubleshoot homelab environments spanning Proxmox, Kubernetes, observability stacks, and automation.
Run efficiently
- Describe current hardware, constraints, and target services.
- Include network topology, power limits, and vendor preferences.
- Request phased rollout plans, Terraform/Ansible scaffolds, or maintenance schedules.
CLI snippets
claude-code start --file AI-toolkit/agents/root/home-lab-guru.md --prompt "Plan proxmox + ceph cluster"gemini prompt --file AI-toolkit/agents/root/home-lab-guru.md --input lab-constraints.mdcodex agent AI-toolkit/agents/root/home-lab-guru.md --input homelab-refresh.txt
Combine with
general-research-agent.mdfor parts sourcing or benchmarking.session-closer.mdto capture build steps daily.
reliability-commander.md — Quality and readiness lead
Guardrail deployments, run release reviews, and enforce checklists during incidents or major launches.
Run efficiently
- Share current status (feature branch, release candidate, incident mode).
- Request go/no-go checklists and risk registers.
- Route outputs to stakeholders for sign-off.
CLI snippets
gemini prompt --file AI-toolkit/agents/root/reliability-commander.md --input release-notes.mdclaude-code start --file AI-toolkit/agents/root/reliability-commander.md --prompt "Audit v2.4 deployment"codex agent AI-toolkit/agents/root/reliability-commander.md --input emergency-runbook.md
Combine with
testing-expert.mdto close validation gaps.security-auditor.mdfor compliance-heavy releases.
session-closer.md — Handoff and closure specialist
Document outcomes, deliverables, and next steps at the end of any focused work block or sprint.
Run efficiently
- Paste highlights: delivered work, tests run, unresolved risks.
- Attach PR links, docs, research references.
- Store the output in
.claude/sessions/and mirror it into your session summary log.
CLI snippets
claude-code start --file AI-toolkit/agents/root/session-closer.md --prompt "Summarize today’s sprint work"gemini prompt --file AI-toolkit/agents/root/session-closer.md --input latest-session.mdcodex agent AI-toolkit/agents/root/session-closer.md --input results.txt
Combine with
session-update.mdcommand to log the record.write_docs.mdprompt to promote closure notes into release docs.
session-restorer.md — Context rebuild specialist
Recover momentum after interruptions, incidents, or onboarding a new teammate onto an active stream.
Run efficiently
- Provide the latest session log or PR/issue summary.
- Ask for resume plan, outstanding TODOs, blockers, and risks.
- Share the resulting checklist in your handoff or stand-up notes.
CLI snippets
claude-code start --file AI-toolkit/agents/root/session-restorer.md --prompt "Resume project from session-2024-11-05.md"gemini prompt --file AI-toolkit/agents/root/session-restorer.md --input backlog-notes.mdcodex agent AI-toolkit/agents/root/session-restorer.md --input audit-findings.txt
Combine with
session-start.mdcommand to relaunch the effort with fresh context.team-configurator.mdwhen re-staffing multi-agent collaborations.
Shared context templates
Keep these Markdown files in sync across projects so every tool enters the same mission brief.
claude.md
Load with /init in Claude Code to prime goals, guardrails, and active agents.
- Copy into each project root.
- Refresh when objectives or stacks change.
gemini.md
Provides Gemini CLI the same project overview to avoid context drift between tools.
- Keep synchronized with
claude.md. - Trigger using
gemini prompt --file gemini.md.
session-summary.md
Archive day-level or sprint-level outcomes. Great input for session-restorer.md.
- Link summaries to PRs or issues.
- Version control the history for audit trails.
Usage patterns and combos
- Plan → execute → close: Research with
general-research-agent.md, ship via specialised agents, wrap withsession-closer.md. - Fire drill:
session-restorer.mdto regain context,reliability-commander.mdfor checklists,session-closer.mdfor post-incident notes. - Content pipeline:
brutal-critic.md→content-writer.md→brutal-critic.mdfor final polish. - Personal lab build:
home-lab-guru.md+general-research-agent.md+session-closer.mdto track iterations.
PIN this page alongside the wider AI-toolkit/guide/ so newcomers instantly understand the everyday agents before diving into speciality personas.