AI Coding Agent Workflow Pack

AI coding agents are fast. Without structure they're also confidently wrong.

They forget your conventions, "fix" tests by weakening assertions, touch 14 files for a one-line bug, and burn your tokens re-reading the same logs. The model is rarely the problem. Usually it's missing context, constraints or verification.

This pack gives you all three, for the three agents indie devs actually use: Claude Code, Cursor Agent and OpenAI Codex.

WHAT YOU GET

• 37-page PDF guide: a practical field manual, not theory

• Template bundle (.zip): every file plain-text and laid out like a repo, ready to copy

• 1-page cheat sheet: Markdown + PDF, to keep next to your terminal

INSIDE THE GUIDE

• Choose the right agent for the job. A scenario-by-scenario comparison (editor-first vs. terminal-first, unattended runs, reviews, CI, monorepos), plus five questions that decide it fast. Qualitative and honest, with no invented benchmark numbers.

• Rule files that actually get followed. Full AGENTS.md, CLAUDE.md and .cursor/rules/*.mdc templates (all four Cursor rule types), plus Codex config.toml and custom agents. Includes the one-source-of-truth layout so your three tools don't drift apart.

• 10 copy-paste workflows, each with the prompt, the steps and the expected output: bug fix (reproduce-first), write tests, safe refactor, code review, docs pass, library migration, feature-from-spec, codebase onboarding, CI/flaky-test triage, performance investigation.

• Multi-agent & subtask decomposition. When splitting helps (and when it just burns tokens), four patterns that work, a five-part subtask contract, Claude Code subagents and skills, Codex custom agents, Cursor worktrees, and headless CI recipes.

• Common failures & token savers. 12 failure modes mapped to cause and fix, and 12 ways to cut token use without hurting quality.

• Security checklist. Secrets, permissions, sandboxing and git safety, with ready-made configs: Claude Code settings.json (allow/ask/deny), a PreToolUse guard hook tested against sample inputs, Cursor cli.json + .cursorignore, and Codex sandbox and env-filtering settings.

WHAT'S IN THE TEMPLATE ZIP

AGENTS.md · CLAUDE.md · .claude/settings.json · .claude/hooks/guard.py · .claude/agents/ (code-reviewer, test-writer, explorer-lite) · .claude/skills/ (/bugfix, /ship-check) · .cursor/rules/ (4 rules) · .cursor/cli.json · .cursorignore · .codex/config.toml · .codex/agents/ (explorer, reviewer) · prompts/ (10 workflows) · checklists/ (security, task brief, decomposition plan, .gitignore lines) · cheat sheet

WHY TRUST IT

• Verified, not vibes. Every file name, settings key and CLI flag was checked against the official Anthropic, OpenAI and Cursor documentation (October 2026). Sources are listed in the appendix.

• Honest comparisons. No fake benchmarks, no "10× productivity" claims.

• Made by a practitioner. Pluto (@Mars_Peakul on X) posts tested AI-agent workflows.

WHO IT'S FOR

• Indie devs and solo founders shipping with Claude Code, Cursor or Codex

• Freelancers who need agents to behave on client codebases

• Small teams that want one shared AGENTS.md instead of three drifting rule files

WHO IT'S NOT FOR

• People looking for a "build an app with one prompt" shortcut

• Teams that need enterprise policy rollout (MDM, SSO). The guide notes where managed settings exist but doesn't cover deployment