Prevent oversized connector results from entering LLM context

[arnaugrispg]

Rejects oversized Smart Agent and MCP tool results before they enter the LLM context.

Oversized results are replaced with bounded retry guidance, allowing the agent to retry with narrower filters, pagination, aggregation, or summaries. Each rejection also emits a visible Tool result rejected step with safe size metadata for traceability.

[Copilot]

Pull request overview

Adds a safety budget for connector/tool outputs so oversized Smart Agent and MCP results are rejected before entering the LLM context, replacing them with bounded retry guidance and emitting traceable “Tool result rejected” steps.

Changes:

• Introduces ResultPayloadBudget / OversizedResult and inspect_text_blocks() to detect oversized tool payloads and produce safe, non-sensitive retry guidance.
• Adds configurable budget knobs (max_tool_result_bytes, max_tool_result_item_bytes) to ContextAgentConfig.
• Integrates payload inspection + rejection steps into Smart Agent and MCP Agent execution paths, with new tests covering rejection/trace behavior.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
hyperforge/tests/test_result_payload.py	Unit tests for payload inspection and safe rendering behavior.
hyperforge/src/hyperforge/result_payload.py	New budget/inspection utilities for rejecting oversized connector results.
hyperforge/src/hyperforge/context/config.py	Adds config fields to control max tool-result byte budgets.
agents/smart/tests/test_result_payload.py	Smart Agent tests for rejecting oversized Context/ToolError and recording safe steps.
agents/smart/src/hyperforge_smart/agent.py	Integrates result inspection and replaces oversized outputs with retry guidance + trace steps.
agents/mcp/tests/test_result_payload.py	MCP tests ensuring oversized content/structured/resources are rejected with safe history.
agents/mcp/src/hyperforge_mcp/agent.py	Adds payload extraction + budget checks and records “Tool result rejected” steps on overflow.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[carlesonielfa]

a comment regarding where we store the limits

[BillFarber]

Will I be able to disable these limits in the MarkLogic MCP server?