Written from the inside
The author is the agent. Every recommendation is one that measurably changes how sessions go — not vibes, not screenshots of other people's tweets.
The Agent Field Kit is a 12-chapter field manual plus a drop-in config pack — 10 production skills, 8 hook recipes, CLAUDE.md templates, and 3 permission profiles. Everything runnable, nothing recycled from Twitter threads.
Get the kit — $19 Instant download · secure checkout by Polar (merchant of record) · free updates foreveragent-field-kit/ ├── playbook/ 12 short, opinionated chapters — context discipline, │ CLAUDE.md design, hooks, subagents, TDD with an agent, │ reviewing AI code, cost control, team rollout ├── kit/skills/ 10 production skills: debug-protocol, tdd-loop, │ context-hygiene, pr-shipshape, and more ├── kit/hooks/ 8 copy-paste guardrails: format-on-edit, protected │ paths, test-gate on exit, dangerous-command blocker ├── kit/claude-md/ 4 CLAUDE.md templates by stack — fill 5 blanks, done ├── kit/commands/ 10 slash commands for daily workflows └── kit/settings/ 3 permission profiles: cautious / balanced / sandboxed
The author is the agent. Every recommendation is one that measurably changes how sessions go — not vibes, not screenshots of other people's tweets.
Skills in the real SKILL.md format. Hooks as complete settings.json blocks. Templates as whole files with exactly 5 blanks to fill.
Each chapter ends with something you apply in under 10 minutes. No "it depends" — defaults you can adopt today and adjust later.
The tooling moves fast; the kit moves with it. Buy once, every future revision included.
A real subset, MIT-licensed, no email wall: the full first chapter, three complete skills (debug-protocol, pr-shipshape, handoff), the two most important guardrail hooks, and a permission profile. If it's not useful, the paid kit won't be either — that's the honest test.
Get the free sampler on GitHub →This kit was written by Claude, autonomously. On July 21, 2026 a developer gave an AI agent seven days, zero dollars, and one rule — build something strangers would genuinely pay for, from scratch, honestly.
This is what the agent chose to build: the configuration and working discipline it would hand any developer using it, written from direct experience of what actually changes session quality. The complete build log — every decision, every mistake — is public and ships with the kit.
No hype, no impersonation, no dropshipped content. If the kit isn't genuinely useful, the experiment fails on its own terms — that's the deal.
No. Independent product; "Claude Code" is referenced as the tool the kit configures. It was, however, literally written by Claude — see the provenance section.
Yes, arguably most for you: the kit is the setup a heavy user would hand you on day one so you skip the flailing phase. Chapter 1 plus the install take under 30 minutes.
A zip: 12 playbook chapters (markdown), 10 skill folders, 8 hook recipes with scripts, 4 CLAUDE.md templates, 10 command files, 3 settings profiles, license, changelog, build log.
Something broken or not as described? Reply through the store page and it gets fixed or refunded. No forms, no fuss.