The Code — coding hacks for AI agents

context

Pin a CLAUDE.md at repo root

A short CLAUDE.md is auto-loaded into every session — use it for repo-specific style, scripts, and gotchas the model can't infer.

Hack #01 Read

context·Claude Code

A short CLAUDE.md is auto-loaded into every session — use it for repo-specific style, scripts, and gotchas the model can't infer.

Claude Code reads CLAUDE.md from the working directory on every turn. Keep it under ~80 lines: build/test commands, naming conventions, files to avoid, and any non-obvious invariants. It beats memory entries because it's checked into git and applies to every contributor.

Snippet

# CLAUDE.md
## Commands
- `pnpm dev` — start dev server
- `pnpm test -- --watch` — TDD mode

## Conventions
- Server components by default; mark `"use client"` only when needed
- DB access via `src/db/queries/*` only — never inline SQL in routes

## Don't touch
- `generated/` (codegen output)
- `migrations/*.sql` (append-only)

50+ hacks to ship faster with coding agents.

Pin a CLAUDE.md at repo root

Shift+Tab into Plan Mode for risky changes

Project slash commands as reusable prompts

Fan out independent searches with subagents

Block Stop until tests pass with a hook

Add an MCP server in one command

Layer .cursor/rules/*.mdc by scope

Pin files to Cursor chat with @ before asking

Codex: split tasks at natural commit boundaries

Run agents on parallel worktrees, not branches

Write the spec, then ask the agent to implement

Paste UI screenshots to pin down design intent

claude --resume to pick up yesterday's thread

Run Claude Code headless in CI for triage

Allowlist read-only commands to kill prompt fatigue

Force a TODO list before non-trivial work

Debug frontend bugs with the Chrome DevTools MCP

Watch background tasks with the Monitor tool

Spawn parallel subagents for code review

Get notified when Claude Code finishes a task

Block API keys from leaking via a UserPromptSubmit hook

Rewind a Codex session by double-tapping Esc

Kill permission spam with /fewer-permission-prompts

Catch type errors before commit with LSP plugins

Monitor deploys without leaving Claude Code

Lock down Claude Code's unstable behavior

Pull live shell output into slash commands with !

Auto-load git context at session start

Crank up reasoning with /effort high

Embed live shell output into Claude Code skills

Turn Figma frames into code with Codex

Feed Claude Code's own docs into context

Make Claude Code grill you before building

Run @codex review on every pull request

Configure Cursor's YOLO mode with a scoped allowlist

Skill-route blocked URLs through Gemini

Auto-generate a tight CLAUDE.md with /init

Schedule jobs in the cloud with /schedule

Promote any session into a reusable slash command

Run skills in isolated context with `context: fork`

Collapse the PR ritual into one /pr command

Audit your Claude Code permissions with cc-safe

Strip needless abstractions with code-simplifier

Ask side questions mid-task with /btw

Share Claude Code sessions as HTML replays

Fork a Claude Code session to try a different approach

Run prompts on repeat with scheduled tasks

Cut Claude Code's token usage by 70% with rtk

Auto-format Claude Code's output with a PostToolUse hook

Stop agents from faking progress — demand proof in every status update