fix(providers): remove default agent tool-call iteration cap

[waleedlatif1]

Summary

• Agent block tool loops no longer stop at a default 20 model round trips; MAX_TOOL_ITERATIONS now caps them only when explicitly set to a positive integer (the execution timeout still bounds a run)
• At the old cap, non-streaming provider loops exited with the model's final turn still requesting tools: those calls never ran and the turn's preamble text came back as the block output, breaking structured-output consumers downstream
• Dropped the /MAX_TOOL_ITERATIONS denominator from iteration log lines, updated the env docs, .env.example, and the test mock's doc (mock keeps a finite test cap)

Type of Change

• Bug fix

Testing

• vitest run providers executor/handlers/agent (2150 passed)
• bun run lint, bun run type-check (apps/sim), bun run check:audits (58 audits), docs-manifest:check
• Full suite left to CI

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing (new tests pass the test-audit authoring gate)
• 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.

Project	Deployment	Actions	Updated
docs	Ready	Preview	Oct 9, 2026 2:18am UTC

[cubic-dev-ai]

No issues found across 12 files

Confidence score: 5/5

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

_{Turn on auto-fix | Re-trigger cubic}

[greptile-apps]

Confidence Score: 4/5

[Medium impact] Changes agent tool-call loop behavior from capped to unbounded by default.

The PR appears safe to merge, with non-blocking documentation and regression-coverage improvements.

Findings

1.  Timeout is not always enabled ▶
2.  Long conversations lack regression coverage ▶

Summary

Removes the default 20-round Agent tool-call limit. MAX_TOOL_ITERATIONS now applies only when set to a positive integer. Provider logs drop the limit denominator, and documentation explains the new default.

• Clarify that self-hosted execution timeouts are also optional.
• Add a regression check for an Agent that finishes after more than 20 tool rounds.
• waleedlatif1 explicitly intends unlimited production loops and a finite shared test limit; neither choice is reported as a defect.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Agent requests a model turn] --> B{Model requests tools?}
  B -->|No| C[Return final answer]
  B -->|Yes| D{Explicit iteration limit reached?}
  D -->|Yes| E[Existing capped-loop behavior]
  D -->|No| F[Run tools]
  F --> A
  G[Configured timeout or cancellation] --> H[Abort execution]
  H -.-> A

Loading

_{Reviews (1) · Last reviewed commit: "fix(providers): remove default agent too..." · Reviewed by Greptile}