User JD on v0.54.0 (beta) reports that the agent reads files but does not launch any subsequent actions — it sits idle without delegating to the code executor. User built out a workflow via code and is new to the product.
Diagnosis from team (21/05/2026)
Agent is reading files but not delegating to the code executor subagent. Root causes identified: (1) user's prompt doesn't explicitly instruct the orchestrator to use the code executor — suggested fix is to be explicit ("Use the code executor to run X"). (2) Model unknown — smaller local Ollama models often skip tool calls and narrate instead; stronger models reliably delegate. (3) User was at 98% of weekly quota — near rate limit may be contributing to clipped agent behavior.
User's intended workflow
After each game session, automatically upload a gameplay video (MP4) to the AI, which then assesses performance against top-500 player benchmarks and surfaces improvement areas based on highlights. Comparable to Omnic AI. This may be a use case that requires clarifying whether OpenHuman currently supports video input/analysis as a first-class capability.
Model and sample prompt not confirmed. No resolution provided. Ticket re-pinged 25/05/2026 with no follow-up.
Reported by: @JDeschaine1985
Discord ticket: https://discord.com/channels/1470856511193616498/1506715564700074005
Priority: medium
User JD on v0.54.0 (beta) reports that the agent reads files but does not launch any subsequent actions — it sits idle without delegating to the code executor. User built out a workflow via code and is new to the product.
Diagnosis from team (21/05/2026)
Agent is reading files but not delegating to the code executor subagent. Root causes identified: (1) user's prompt doesn't explicitly instruct the orchestrator to use the code executor — suggested fix is to be explicit ("Use the code executor to run X"). (2) Model unknown — smaller local Ollama models often skip tool calls and narrate instead; stronger models reliably delegate. (3) User was at 98% of weekly quota — near rate limit may be contributing to clipped agent behavior.
User's intended workflow
After each game session, automatically upload a gameplay video (MP4) to the AI, which then assesses performance against top-500 player benchmarks and surfaces improvement areas based on highlights. Comparable to Omnic AI. This may be a use case that requires clarifying whether OpenHuman currently supports video input/analysis as a first-class capability.
Model and sample prompt not confirmed. No resolution provided. Ticket re-pinged 25/05/2026 with no follow-up.
Reported by: @JDeschaine1985
Discord ticket: https://discord.com/channels/1470856511193616498/1506715564700074005
Priority: medium