docs: add Qwen long-flow troubleshooting

[mason5052]

Summary

• Adds a Qwen3.5-27B-FP8 troubleshooting note for automation flows that run for hours or repeat commands.
• Points users to existing execution monitoring, planning, and tool-call limit controls before assuming a runtime bug.
• Adds a cross-reference from the advanced flow execution docs to the Qwen guide.

Problem

Issue #344 reports a Tomato VulnHub penetration test running for more than 5 hours with Qwen3.5-27b through an OpenAI-compatible custom provider. The flow appeared to repeat commands and make slow progress. PentAGI already documents execution monitoring, task planning, reflector recovery, and hard tool-call limits, but the Qwen guide did not connect those controls to long-running local/custom model flows.

Refs #344

Solution

• Document how to diagnose long-running or looping Qwen/custom-model flows using existing controls: EXECUTION_MONITOR_ENABLED, EXECUTION_MONITOR_SAME_TOOL_LIMIT, EXECUTION_MONITOR_TOTAL_TOOL_LIMIT, AGENT_PLANNING_STEP_ENABLED, MAX_GENERAL_AGENT_TOOL_CALLS, and MAX_LIMITED_AGENT_TOOL_CALLS.
• Recommend bounded prompts with target, scope, expected output, and stopping criteria instead of broad requests such as only performing penetration testing on a host.
• Tell operators to compare PentAGI flow logs, Docker logs, and provider/vLLM logs to separate model weakness, tool loops, target complexity, and provider/tool-call issues.

User Impact

Users running local or custom Qwen3.5-27B-style models get a practical first troubleshooting path without changing PentAGI runtime behavior, defaults, dependencies, or generated artifacts.

Test Plan

• git fetch upstream main
• git diff --check upstream/main...HEAD
• git diff --name-only upstream/main...HEAD
• Verified all referenced env vars exist in .env.example and backend/docs/config.md.
• No automated tests were run because this is a docs-only change.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds troubleshooting guidance for runaway/looping automation flows in the vLLM Qwen3.5-27B-FP8 guide and cross-links it from the execution monitoring documentation to help users tune monitoring, prompt scope, and diagnostics.

Changes:

• Added a new troubleshooting subsection for long-running/repeating automation flows (prompt bounding + logging checklist).
• Linked the new troubleshooting section from flow_execution.md to make it discoverable from execution monitoring docs.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
examples/guides/vllm-qwen35-27b-fp8.md	Adds a new troubleshooting entry focused on long-running/repeating flows and mitigation steps.
backend/docs/flow_execution.md	Adds a documentation cross-link to the new troubleshooting section for Qwen-style local/custom deployments.

---

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