riffpad: Steer AI Coding Agents From Your Phone

Where this fits

You're mid-refactor, the agent is grinding through a long task, and you need to step away from your desk—but you don't want to kill the session or blindly trust the output. With riffpad, you can keep an eye on the agent's progress, approve sensitive actions, and nudge it in the right direction from your phone, while the code stays on your machine with your API keys.

Why agents benefit

• Remote supervision: The agent can run long sessions unattended, but the human stays in the loop via push notifications and a live view—no more babysitting a terminal.
• Context-aware approvals: Approve or reject tool calls (file writes, shell commands) from the phone, so risky steps wait for your sign-off even when you're away.
• Steering on the fly: Send new instructions or clarifications mid-session without interrupting the agent's flow or losing its current context.
• Multi-agent support: Works with Claude Code, Codex, and Kimi, so you can manage different agents from one phone interface.
• Privacy-preserving bridge: A local daemon connects to your phone through a zero-knowledge relay—your code and API keys never leave your machine.

Practical scenarios

Deploying hotfixes from the train

Your agent is fixing a production bug. You're commuting. Instead of waiting to get home, you watch the agent's diffs, approve the final commit, and let it push—all from the Riffpad app.

Long-running refactors with checkpoint reviews

A large refactor touches many files. You set the agent running and check in periodically. When it hits a risky step (e.g., deleting a module), you get a notification and approve only after reviewing the diff on your phone.

Pairing with a junior developer's machine

You're mentoring remotely. The junior runs an agent locally, and you join as a viewer to guide it—steering prompts and catching mistakes in real time, without needing screen-sharing or SSH.

Add it to your agent workflow

Install the skill into your agent's skill directory:

npx skills add riffpad/riffpad --skill riffpad

The skill will guide the agent through the full setup: installing the daemon (riffpad install), logging in, and pairing your phone. After that, you can invoke it with natural language like "start riffpad and show me the current session status." The skill handles the onboarding flow step by step, so the agent knows exactly what to run.

Compare before adopting

• Quality signals: Check the repo's star count (126 at time of writing), recent commit activity, and whether issues are being resolved. A small but active project can be more reliable than a large, dormant one.
• Maintenance freshness: Look for updates that track changes in Claude Code, Codex, or Kimi's CLI interfaces—agent tooling shifts fast.
• Alternatives: Consider cloud IDEs, tmux + SSH, or simple notification hooks. riffpad's advantage is the zero-knowledge relay and native phone UX, but those may not matter if you only need basic alerts.
• Workflow fit: Verify that the agent you use is supported and that the approval flow matches your risk tolerance. Test with a low-stakes session first.

Why it is worth tracking

riffpad solves a real pain point—keeping a human in the loop without being tied to a desk—and its design (local daemon, zero-knowledge relay) aligns with privacy-conscious development. The project is young but gaining traction, and the skill's existence means your agent can set it up without manual reading. Keep it on your radar: if you regularly run long agent sessions or work across devices, this could become a core part of your workflow. For a deeper look, see the skill page on OpenAgentSkill for community notes and version history.