Validate loop depends_on entries against loop subgraphs

[zhongkaifu]

Motivation

• Prevent invalid workflows where a loop node declares depends_on entries that are actually nodes inside its own body_subgraph, and provide LLM-facing guidance to resolve naming conflicts.

Description

• Add helper _build_loop_depends_on_subgraph_conflict that detects overlap between a loop's depends_on and its sub_graph_nodes and returns a structured payload including loop_id, depends_on, sub_graph_nodes, overlap_nodes, and an LLM suggestion in requirements_suggestions.
• Enforce this check in add_loop_node using the provided sub_graph_nodes during creation and return a needs_more_work payload if overlap exists.
• Enforce this check in update_loop_node by computing effective depends_on (existing or provided) and effective subgraph nodes (current body nodes or provided), validating the combined sets before applying updates.
• Hardened retrieval of existing depends_on by ensuring it is a list before using it in the conflict check.

Testing

• No automated tests were run for this change.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d0bce800f5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Validate against existing loop body nodes

When update_loop_node receives sub_graph_nodes, the conflict check only uses normalized_nodes, but the update path later only attaches nodes (it never detaches existing body nodes). That means the effective body is still the existing current_body_nodes plus the newly attached ones. If a caller updates depends_on while adding new sub_graph_nodes, depends_on can still reference an already-attached body node and the check will pass, leaving a loop that depends on its own body. Consider validating against the union of current_body_nodes and normalized_nodes (or recompute after attaching) so existing body nodes are included.

Useful? React with 👍 / 👎.