The best coding agent features, according to hackers
10 coding agent features recommended in Hacker News comments over the past year, ranked by how many separate people recommend each one.
All coding agent features
- Claude Code plan modeUsed to review and steer an agent's plan before it edits code and to understand changes; the common caveat is clunky UX, and newer models need it less.41recommend
- Claude Code auto modePeople recommend it as a safer, less interruptive alternative to skipping permissions, since a classifier approves routine commands. Critics say it is not a sandbox or security guarantee and its classifier misses some risky actions.20recommend
- Claude Code RoutinesPeople recommend it for running Claude Code on a timer in Anthropic's cloud, replacing self-hosted cron jobs; the common caveat is unreliability and vendor lock-in.16recommend
- Claude Code Dynamic WorkflowsLets Claude Code run many parallel subagents from one orchestrator, quick to set up and good on complex long tasks. Main caveat: it uses many tokens and is expensive.13recommend
- Claude Code UltracodePeople recommend it for unattended, fast coding work and for finding bugs in large codebases. The most common caveat is that it burns through tokens and usage quotas quickly, and some find it unreliable where correctness matters.12recommend
- Claude Code ChannelsLets an outside integration hook into a running Claude Code session without restarting or rebuilding context. The most common caveat is that it still feels rough, with permission prompts and buggy behavior.11recommend
- Claude Code /advisorLets a model like Sonnet hand hard steps to Opus in a clean sub-agent context. One critic notes it burns through token budgets.8recommend
- Claude Code AskUserQuestionLets Claude ask multiple-choice questions instead of plain text, speeding decisions. Main caveat: critics say it was turned on by default without notice and is hard to disable.8recommend
- Claude Code output stylesRecommended for embedding custom instructions in the system prompt, with periodic reminders. Critics say it does not reliably fix model behavior.6recommend
- Claude Code scheduled tasksPeople like that it automates recurring prompts and reduces manual work through a simple UI. The common caveats are that it is limited to a few cloud tasks and falls short of cron jobs.6recommend