bug: script in skills not run in subsequent same-chat interactions

Description

I have an instance of computer running with AzureOpenAI connection to an LLM. Skills to produce an artifact (e.g. .docx file) are provided as a set of markdown files like:

.agents/skills/SKILL.md # <-- here references to the references/*.md + check_delivery.py
.agents/skills/references/artifact-delivery.md
.agents/skills/check_delivery.py

The script check_delivery.py is referenced in SKILL.md and everywhere else in the prompt files within .agents/skills/references because I wanna be sure it is executed anytime the artifact is completed.

The observed behaviour is that the correct workflow is executed at the first chat interaction e.g. "produce a word document with this text ...". In a subsequent message, e.g. "refine the chapter ... in the previously generated docx" it fails to execute check_delivery.py.

Steps to reproduce

• Connect computer to an LLM
• Write a skill that references a script to be executed at the end of the workflow
• Store skills (markdown + script) in .agents/skills
• Start a conversation that makes the agent use the skill and the script
• Iterate with a second message in the same chat
• Check if the script is run

Suggested behavior

• The second interaction should follow the same workflow as the first one thus having the script run at the end both times.

Start with the skill workflow described in .agents/skills/SKILL.md and the referenced .agents/skills/check_delivery.py, then compare execution during the first and subsequent messages in one chat. Done means the referenced script runs at the end of the workflow for both the initial request and a refinement request.