[Fleet Execution] Implement Tier 5 Advanced Custom AI Workflows

[jules-fleet]

Goal Reference

This issue was generated from .fleet/goals/sdk-examples.md. The worker agent should read this file for full context including verification commands, constraints, and structural guidance.

Objective

Implement Tier 5 advanced custom AI workflows, specifically demonstrating Vercel AI SDK tool filtering (stitchTools) and a hybrid CI visual testing agent skill.

Code-Level Diagnosis

Code path: packages/sdk/examples/
Mechanism: The SDK exposes an AI SDK adapter (stitchTools from dist/src/ai.js), but there is no practical example demonstrating how to filter these tools. Additionally, CI visual testing strategies and visual regression evaluation guides are missing.
Root cause: Advanced integrations targeting AI framework adapters and CI pipeline visual testing lack concrete example implementations.

Current Implementation

// Missing examples for AI tool filtering and CI visual regression handling

Proposed Implementation

Files to modify: Scaffold directories for tool-filtering and ci-visual-testing. Implement tool-filtering/index.ts to showcase stitchTools({ include: [...] }) usage. Provide a SKILL.md and a ci-runner.ts script for ci-visual-testing to handle visual regressions.

Integration (Before -> After)

+ // Advanced AI SDK filtering and CI runner logic added to expand the examples coverage

Test Scenarios

1. Tool Filtering: The script correctly accesses the AI adapter and limits available operations.
2. CI testing context: The ci-runner.ts properly implements deterministic baseline regeneration and comparison points.

Target Files

• packages/sdk/examples/tool-filtering/README.md
• packages/sdk/examples/tool-filtering/index.ts
• packages/sdk/examples/ci-visual-testing/SKILL.md
• packages/sdk/examples/ci-visual-testing/scripts/ci-runner.ts
• packages/sdk/examples/ci-visual-testing/README.md
• packages/sdk/README.md

Boundary Rules

Restrict your modifications exclusively to the files listed in the Target Files section. Ensure your source changes are entirely backward-compatible if unowned tests outside your boundary fail. Retain all existing file names and locations outside your explicitly declared target list.

---

Fleet Context

• Source: jules:session:13642316243044202608
• Link: https://jules.google.com/session/13642316243044202608