@yusukeshib/pi-ping0.1.0

pi-ping

An opt-in Pi extension that sends periodic wake-up messages when the agent is idle, starting another agent turn. Disabled by default, with a 5-minute default interval.

Installation

After the package is published:

pi install npm:@yusukeshib/pi-ping

For local development, install the checkout instead:

pi install /path/to/pi-ping

Run /reload in Pi after installation. Tested with Pi 0.99.2.

Commands

Durations accept s (seconds), m (minutes), and h (hours). Bare numbers are minutes: 5 means 5 minutes. Fractional values such as 0.5m work too. The range is 1 second to 24 hours; 5 minutes is recommended for normal use.

Behavior

• Waits a full interval after enabling or after agent_settled, when Pi has finished all automatic work. It does not wake on agent_end, so it does not interrupt Pi's built-in retries or automatic compaction.
• Does not send while the agent is busy, messages are queued, an extension dialog is open, or the TUI editor contains a draft. Checks again after another interval.
• Displays a distinct pi-ping custom message rather than impersonating user input. The message asks the agent to continue unfinished authorized work or retry after a transient network failure. It explicitly discourages inventing work, repeating completed side effects, bypassing approvals, or ignoring a request to stop.
• Uses at most one timer. Cancels it when a run starts and restarts it after settlement. Disable, shutdown, session replacement, and reload clean up the old timer.
• Saves enabled state and interval on the session's active branch. Resume, reload, and forks that inherit the saved state restore those settings. New sessions start disabled. Navigating with /tree restores the destination branch's settings.

Important limitations

While enabled, pi-ping starts model turns even after work is complete or after an Esc interruption. This can consume tokens and incur charges. Use /ping disable to stop it.

This extension does not restore network connectivity. If a network failure ends Pi's run and leaves it idle, the extension attempts another model turn after the configured interval. If the network is still unavailable, that turn can fail again; another check follows after the failed run settles.

It cannot recover an exited Pi process, wake a sleeping operating system, unblock a frozen event loop, or interrupt a hung run that never returns to idle. It is not a watchdog that forcibly aborts operations. Whether work actually resumes depends on the conversation, permissions, model, and connectivity.

Development and validation

npm ci --ignore-scripts
npm run check
npm run smoke
npm pack --dry-run

check runs TypeScript checks and 10 fake-timer tests. smoke requires an installed pi command and Python 3. It runs Pi's actual RPC lifecycle with isolated settings and a deterministic local provider, verifying simulated network failure, wake-up, recovery, and disable. It makes two local mock calls, no external model calls, and incurs no API charges. Actual network outages and long-running real-model continuation have not been tested.

Publishing to npm

The initial release is 0.1.0. Do not bump the version before the first publication.

npm run release:dry-run
npm run release

release publishes to npm. The prepublishOnly hook runs type checks and tests first. publishConfig selects the official npm registry and public access. Follow npm's CLI instructions if authentication or two-factor verification is required. Run npm run smoke separately before publishing when needed.

Only extensions/, README.md, LICENSE, and package.json are included in the package. After the initial publication, subsequent releases require a new version: npm does not allow republishing the same version.

Uninstallation

pi remove npm:@yusukeshib/pi-ping

For a local checkout, use pi remove /path/to/pi-ping instead. Then run /reload.