claude-workflow

Use plan mode when the approach is open, the change is architectural, or a wrong direction would be costly to redo. If the plan stops matching what you find during implementation, re-plan instead of patching forward.

Run /compact deliberately at natural task boundaries, such as after exploration or before starting a new subtask, rather than relying on auto-compaction. Avoid compacting mid-implementation or while debugging, since you need that working context.

Use subagents liberally to keep the main context window clean. Offload research, exploration, and parallel analysis to subagents, assign one task per subagent, and throw more compute at complex problems via additional subagents.

Add a PreToolUse hook in .claude/settings.json that matches Edit and Write operations and blocks changes to protected config files like .eslintrc, prettier.config.js, and tsconfig.json. The hook forces fixing the code instead of weakening the config.

Running prettier and tsc after every edit is slow and repetitive. A PostToolUse hook can accumulate edited file paths, and a Stop hook runs prettier --write and tsc --noEmit once on the deduplicated list when Claude pauses.

High-risk changes involving auth, data, or infrastructure should go through cross-model adversarial review before shipping. The workflow recommends running /codex:adversarial-review via the codex-review skill for these cases.