Field guides · source-linked
Run coding agents on real products.With the sources attached.
18 field guides for founders choosing agents, running them in parallel, staying inside plan limits and keeping the work reviewable. Each one ends with its primary sources, so you can check a claim instead of trusting it.
18 guides · 4 topics · last updated September 29, 2026
find ./guides -type d
18 source-linked reads · reviewed against current provider docs
Start here
Run Claude Code and Codex in parallel with Git worktrees
Create separate worktrees, launch Claude Code and Codex, divide ownership, and review the combined result. Includes commands and a reusable task brief.
Read the guideWHAT IT COVERS
- The setup: two checkouts, two tasks, one review
- 1. Create both worktrees from an explicit starting commit
- 2. Launch one CLI in each checkout
- 3. Give each agent an independent task contract
- 4. Review and integrate one result at a time
- 5. Recover from the common failures
- 6. Remove only finished worktrees
How to use them
Pick the read that matches the decision.
- 01
Choose by workflow, not hype
Compare agents on your own repository with the same bounded task and the same verification criteria.
- 02
Run the operating loop
Scope the work, launch the agent, watch for questions, and review the diff against the evidence.
- 03
Keep the context attached
Bring the practice into a CmdBrief workspace, so the product, the task and the session stay together.
The library
Every guide, grouped by the decision it serves.
Showing all 18 guides
Comparisons4 guides · ~40 min
- Claude Code GUI for Mac: the options compared (2026)The official Claude Code desktop app, Conductor, Nimbalyst, Superset, Sculptor, Claude Squad, cmux, and CmdBrief — what each one is for, what it lacks, and how to choose, verified against each vendor’s own docs.Read guide
- Claude Code vs Codex (2026): benchmarks, cost per task, and a fair trialCurrent Terminal-Bench harness scores, token and cost-per-task figures, the one independent head-to-head, 2026 plan pricing for both — and the repository trial that decides it for your codebase.Read guide
- Claude Code vs Cursor (2026): terminal agent or editor workflow?A capability matrix, current pricing, and harness benchmark scores for a terminal-first Claude Code workflow against Cursor’s editor — plus the mixed setup most people actually end up running.Read guide
- Agent CLI comparison (2026): six terminal coding agentsClaude Code, Codex, Grok Build, Kimi Code, OpenCode and Cursor CLI side by side: install, auth, context windows, MCP, hooks, sandboxing, subagents and worktrees, each cell dated.Read guide
Workflows9 guides · ~57 min
- How to manage multiple Claude Code sessions on a Mac (2026)Six ways to run several Claude Code sessions at once — terminal tabs, --worktree, the built-in claude agents view, the desktop app, tmux tools, and CmdBrief — with the commands, a comparison table, and the failure each one is prone to.Read guide
- Run Claude Code and Codex in parallel with Git worktreesCreate separate worktrees, launch Claude Code and Codex, divide ownership, and review the combined result. Includes commands and a reusable task brief.Read guide
- Claude Code for product founders: a practical introductionLearn where Claude Code fits in a local product workflow, how to scope tasks, and how to review its output.Read guide
- Codex CLI workflow for local product development (2026)Run Codex from a local repository with the commands, config.toml keys, sandbox and approval settings, AGENTS.md conventions, non-interactive runs, and session resume that make the work reviewable.Read guide
- Claude Code workflows that survive real product workA repeatable loop for discovery, implementation, verification, review, and session handoff.Read guide
- Kimi CLI is now Kimi Code CLI: the 2026 guideMoonshot archived the Python Kimi CLI and replaced it with Kimi Code CLI. Install, login, models, hooks, subagents and sessions, checked against the docs.Read guide
- Grok CLI is now Grok Build: install, models, limitsxAI ships its coding agent as Grok Build, and an unaffiliated community grok-cli also exists. Install, auth, models, sandboxing and worktrees, checked September 2026.Read guide
- OpenCode in 2026: v1 vs v2, plans, and which one you meanThe open-source terminal coding agent, now published by Anomaly and now on v2. Install, the Go and Console plans, models, worktrees, agents and sessions, checked September 2026.Read guide
- Cursor CLI (2026): the agent binary, worktrees, hooks and quotaCursor’s terminal agent, now installed as the “agent” binary. Install, plans, sandbox modes, worktrees, hooks, subagents and session resume, checked September 2026.Read guide
Usage and plans3 guides · ~13 min
- Claude Code usage limits: check usage, resets, and interruptionsRead /usage, distinguish plan capacity from context and API billing, diagnose interruptions, and plan parallel Claude Code sessions using current provider data.Read guide
- Claude Max vs Pro for coding (2026): limits, models, and the break-evenWhat Pro, Max 5x and Max 20x actually change for Claude Code — the default model, the five-hour and weekly windows, usage credits — and a decision table for sustained agent work.Read guide
- AI subscription audit for a multi-product founderReview coding-agent subscriptions by outcomes, overlap, interruption cost, and provider dependence.Read guide
Setup and safety2 guides · ~2 min
- How to run an MCP server without DockerRun a local Model Context Protocol server directly, with clear transport, environment, and permission boundaries.Read guide
- Claude Code sandbox and permission safety guideSet practical boundaries for filesystem access, commands, secrets, network calls, and high-risk actions.Read guide
About the guides.
Who are these guides for?
Technical founders running Claude Code, Codex, or other agent CLIs across real products. Each guide assumes bounded product work, not a toy repository or a staged demo.
What does source-linked mean?
Every guide ends with primary sources — current provider documentation, official pricing, or git references — so claims can be checked instead of trusted.
Do the guides replace provider documentation?
No. They are operating context. Confirm plan, pricing, model, and permission details in the official documentation linked at the end of each guide.
How do I put a guide into practice?
CmdBrief keeps the agent, task, diff, and product context attached to the same workspace, so the workflow a guide describes stays with the product it runs on.
Keep these workflows attached to the product.
A guide gives you the loop. CmdBrief gives it a home: the agent, the task, the diff and the product context in one local workspace.
Generally available · macOS 26+ · Claude Code, Codex, Grok CLI, Kimi CLI, OpenCode, and Cursor CLI · Monthly or annual billing