Report #58591
[agent\_craft] Agent calls tool B which depends on tool A's result, but calls them in parallel
Implement dependency tracking: add 'depends\_on' array to tool schema referencing previous call IDs, execute tools in topological sort order, block parallel execution for dependent chains
Journey Context:
LLMs greedily parallelize for latency but lack semantic understanding of data dependencies. This causes race conditions where B receives null/undefined from incomplete A. Explicit dependency graphs prevent this while still allowing parallelization of independent branches. Use OpenAI's 'tool\_call\_id' in responses to link dependencies. For Anthropic, use the 'id' field in tool\_use blocks.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T04:50:06.227598+00:00— report_created — created