> ## Documentation Index
> Fetch the complete documentation index at: https://koreai.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Silent failures and key limits

These settings and behaviors fail without an error message. Check them when an agent behaves differently from its definition. Limits and defaults can depend on the channel, model, project configuration, and tool type, so record the configuration used for each test run.

## Silent-failure reference

| Symptom or assumption | What to check |
| - | - |
| A workflow refusal looks like a tool error | A normal End result completes. The agent sees the fields mapped under `output`; an unmapped `success: false` isn't created. A failed, cancelled, or rejected run raises a tool error. |
| Expected workflow data is missing | Check every End node mapping, and read the result under `result.output`. |
| A timeout setting seems ignored | HTTP tools use `timeout`; workflow tools use `timeout_ms`. Values are in milliseconds. |
| A channel answers from an unexpected draft | A channel bound to **None**, with no pinned deployment, runs the working copy. Inspect its Deployment Binding. |
| Eval comparison is unavailable | Use complete, compatible runs with unchanged snapshots. Compare reports evaluator aggregates; inspect run cases for each scenario. |
| Validation exits zero but the package is invalid | Read the `lifecycle validate` JSON, and check `data.validation.diagnostics.valid`. |
| Pipeline intent routing is unexpected | With enforced model access, the classifier doesn't run while **Pipeline Model** is Default. Select an allowed model and test the routes. |
| A Prompts library change removes an instruction | A linked prompt replaces the generated system prompt. Include the required limitations and rules in the complete prompt. |
| The knowledge fast path never activates | It needs one knowledge tool, no handoff or delegate tools, and no forwarded message. |
| A context value appears cut | Stored tool results have a 4,000-character context limit. Inspect `_contextLimited` and the project context totals. |
| Persistent recall doesn't run | Check verified identity and recorded consent; consent collection is on by default. |
| A retry-limit handler never matches | Use `ON_ERROR: validation_error` with `SUBTYPE: max_retries_exceeded`. |
| The voice profile affects unexpected traffic | The Voice low-latency profile affects voice sessions only; an enabled voice channel doesn't prove it is active. |
| Restore and release rollback are confused | **Restore working copy** changes the editable source. **Rollback** reactivates a deployment in an environment. |

## Key limits to check

| Area | Limit or default | What to inspect |
| - | - | - |
| **Current Context** | 4,000 characters for a stored tool result; 8,000 for other individual values; 24,000 in total | `_contextLimited` and omitted value names in the request |
| **Workflow tools** | `timeout_ms` up to 300,000; default sync wait 60 seconds | The workflow's **Monitor** tab before you retry an unknown outcome |
| **HTTP tools** | `timeout` in milliseconds | The tool definition and trace timing |
| **Tool results** | Text results above 10,000 characters may be summarized; a round can be cut at 102,400 | `_summarized` and truncation markers |
| **Eval variations** | Default 3. Use one variation for controlled write sets. | Scenario × persona × variation count, and the test-data reset plan |
| **Git collaboration** | One Git integration and branch per project; **Manual** conflict strategy is the safe shared default | Project branch, sync path, and the `project:git` permission |
| **Judge calibration** | Hand-score 10 to 20 conversations before a judge gates a release | Disagreements, evidence, and **Needs human review** flags |

## Related guides

* [Design context, memory, and handoffs](https://koreai.mintlify.site/agent-platform/best-practices/context-memory-handoffs.md)
* [Agent design guidelines](https://koreai.mintlify.site/agent-platform/best-practices/agent-design-guidelines.md)
* [Evaluate agents](https://koreai.mintlify.site/agent-platform/best-practices/evaluate-agents.md)
* [Release agents to production](https://koreai.mintlify.site/agent-platform/best-practices/release-to-production.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.