fix(runtime): recover reliably from stalled Penguin sessions

[Maximooch]

Summary

• add bounded provider stream watchdogs, retry budgets, typed terminal failures, and durable continuation state so stalled requests recover without requiring a blind manual resume
• make request ownership, busy/idle status, heartbeat, SSE replay, and TUI hydration resilient to disconnects, cancellation, auxiliary-hook failures, and concurrent requests
• preserve native tool-call adjacency and harden checkpoint/storage workers so background persistence cannot wedge the reasoning loop
• improve prompt/cache behavior and active-turn prompting while preserving PR feat: add session-scoped /goal and /247 commands #72's uncapped long-lived /goal runs
• keep delegated helper loops bounded independently of the primary goal loop
• add runtime diagnostics, benchmark/evidence artifacts, and focused fault-injection/state-machine coverage

PR #72 integration

This branch includes the merged session-goal work from #72. Primary goal runs remain uncapped unless an explicit iteration limit is supplied. Individual provider waits/retries and delegated helper loops remain bounded so one stalled operation cannot block the session indefinitely.

Verification

• uv run pytest tests -q: 2449 passed, 3 skipped, 113 deselected
• bun run typecheck: passed
• targeted Penguin TUI integration suite: 59 passed
• full TUI suite: 1101 passed, with 9 process-startup tests hitting the default 5s timeout under the full concurrent run; all 9 passed when their three files were rerun with --timeout 15000 (81 passed)
• git diff --check: passed

Scope

CWM v2 is explicitly deferred to a separate follow-up PR so its context-selection behavior can be evaluated independently from these runtime recovery changes.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
penguin	Ready	Preview, Comment	Jul 13, 2026 9:08pm

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 9c9267b9-9bbd-4190-8524-3cd709002a02

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/runtime-reliability-recovery

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}