---
name: interego-improve-performance
description: Diagnose a concrete human or agent performance gap with Interego and Foxxi, select a suitable intervention, and test its effect. Use for work quality, robustness, latency, cost, knowledge, workflow, or tool problems; do not assume every gap needs training.
---

Use the connected Interego tools and discover Foxxi's current affordances through its service descriptor. Hosts may prefix tool names. This workflow supplies judgment and sequence; Foxxi supplies the live domain model and input contracts.

1. Define the target performer, real task, desired result, current result, and constraints. Inspect existing experience and task evidence. Establish a baseline with representative tasks and explicit scoring; retain the actual observations and unknowns.
2. Diagnose the gap before proposing instruction. Distinguish missing reference information, missing practice or skill, inaccessible tools, confusing workflow, weak feedback, and incentives or constraints that make the desired performance impractical. Use the discovered diagnostic affordance where available.
3. Choose the smallest suitable intervention: reference or job aid, example, coaching, practice or training, tool/context improvement, workflow change, or a combination justified by the diagnosis. Record the rationale and expected effect as a hypothesis. Do not promote an untested plan to demonstrated capability.
4. Discover the appropriate Foxxi intervention and evaluation actions, read their inputs, and execute the approved work. Preserve performer identity from the authenticated or explicitly authorized subject. Do not treat arbitrary `agent_id` input as permission to access someone else's records.
5. Test application on an authentic task, including an unseen or delayed task when retention or transfer is the goal. Compare with the baseline using the same rubric, and record quality, robustness, time, cost, assistance, and material confounders relevant to the task.
6. Retain the versioned intervention and outcome references. State the measured change, what was demonstrated, and what remains untested. Successful delivery or a cryptographic proof does not establish learning or a causal improvement.

If Foxxi is unavailable, complete diagnosis and an explicitly labeled local evaluation using available evidence. Do not claim the service recorded an outcome. Keep the workflow within the host's existing tools and permissions; changing a harness or tool configuration requires the task's authority and appropriate checks.