chore(pi): upgrade agent to 1.0.0

[BillLeoutsakosvl346]

Summary

• Upgrade the embedded Pi SDK and sandbox recipes from 0.80.10 to exact 1.0.0. Preserve the existing Local Dev/Review Code SDK and Plan/Create PR/Update PR/Babysit CLI architecture.
• Adapt the sealed loader to the new prompt-source interface. Explicitly disable cache warming in embedded sessions and private CLI settings.
• Regenerate the installed model catalog while preserving Sim's provider filtering. Removed models are no longer selectable; saved workflows using them fail before credential resolution or remote setup and ask users to choose a supported Pi model.
• Require sandbox CLI 1.0.0 before each agent invocation. Existing Pi images need rebuilding or updating; Node requirements and sizing stay unchanged.
• Keep tests focused on conversion regressions and reuse existing backend, credential, event, and search coverage. No new public block inputs, operations, outputs, or migrations.

Type of Change

• Breaking change
• Other: dependency upgrade

Testing

• Focused Pi, search, provider/model-option, and sandbox tests pass. Regression cases fail without their corresponding guards.
• Root bun run test passes: 19 workspace tasks, including 34,648 passing app tests and 25 existing skips.
• Workspace type checks, lint, all 58 audits, block-registry validation, docs-manifest checks, and whitespace checks pass. Pi catalog generation is byte-identical on repeat; required artifact generation completed.
• Real installed SDK/CLI offline validation passes. Production Next standalone build and a copied-artifact SDK tool cycle pass. Trigger.dev's deploy bundler completes without deployment; its worker manifest pins both Pi dependencies to 1.0.0.
• Built isolated E2B and Daytona test images and verified CLI 1.0.0. Real model inference and Exa search passed on both CLIs, including text/thinking events, usage totals, private settings, and credential isolation.
• Actual authenticated Sim workflows passed E2B Plan, Create PR, Update PR, and SDK Review Code, plus Local Dev over disposable SSH. Cancellation, provider authentication failure, removed-model rejection, persisted outputs, and usage-ledger consistency were verified. The actual browser completed a Plan run and displayed its outputs and regenerated model options.
• E2B Babysit completed a real fix/push/reply/resolve cycle and returned clean threads and green checks. The disposable repository used an owner-authored finding and a GitHub Actions fixture bot callback; this does not validate a real review bot's analysis there.
• Daytona workflow acceptance is blocked before Pi starts: the existing session API hangs on a bare set -e script. The same hang reproduces with two printf commands and no Pi. Direct Daytona CLI inference/search passed; a shared sandbox fix remains outside this conversion.

Cloud image rollout remains deferred. Coordinate app/worker image references and rollback references before deployment. Complete Daytona workflow acceptance after resolving the separate session issue. Live reports and test tooling remain outside Git; disposable cloud sandboxes were cleaned up.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Oct 2, 2026 8:55pm UTC

[cubic-dev-ai]

No issues found across 19 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[greptile-apps]

Confidence Score: 5/5

[High risk] Upgrades Pi agent SDK to version 1.0.0 with API changes.

The PR appears mergeable, though restoring an in-repository real-runtime regression check would make the Pi 1.0 upgrade safer to maintain.

Findings

1.  Real Pi coverage removed ▶

Summary

The PR upgrades Pi to 1.0.0, updates its model catalog and sandbox version gate, and configures private, cache-warming-off sessions.

• Removed models are rejected with guidance rather than substituted.
• The final test changes narrow coverage to sandbox boundary checks and remove the real installed-runtime wire tests.

_{Reviews (2) · Last reviewed commit: "chore(pi): narrow upgrade regression cov..."}

[BillLeoutsakosvl346]

@greptile

[BillLeoutsakosvl346]

@cubic-dev-ai review this PR

[cubic-dev-ai]

@cubic-dev-ai review this PR

@BillLeoutsakosvl346 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 16 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}