What kind of root causes can it identify?

It classifies issues into four categories: knowledge gaps, missing documentation, incorrect instructions, or missing runtime context.

What files does the skill require to function?

It requires access to a session log folder containing conversation_transcript.jsonl and .issue.yml files.

What is the primary purpose of the investigate-issues skill?

It is designed to analyze session logs from a LearningAgent to determine the specific reasons why an agent failed a task or encountered an issue.

How does it prove its findings?

The skill is instructed to be specific about evidence by referencing exact line numbers from the conversation transcript in its investigation report.

Does this skill automatically fix the agent's behavior?

No, this skill focuses strictly on investigation and reporting. A separate incorporation step is required to update the agent's knowledge base or instructions based on these reports.

Learning Agent Issue Investigator

Name: Learning Agent Issue Investigator
Author: Unsupervisedcom

byUnsupervisedcom

•

분석 및 모니터링

Analyzes AI agent session logs to identify root causes of performance issues and generates detailed investigation reports.

The investigate-issues skill automates the diagnostic phase of the LearningAgent workflow by systematically reviewing conversation transcripts and session logs. It helps developers understand why an agent failed or underperformed by categorizing issues into specific root causes such as knowledge gaps, documentation deficits, or instruction errors. By providing evidence-based reports linked to specific transcript line numbers, this skill enables a structured approach to iteratively improving AI agent reliability and accuracy.

주요 기능

01Automated transcript analysis and keyword-based log searching

02Standardized root cause taxonomy for diagnostic consistency

03Bulk processing of multiple identified session issues

04Evidence-based reporting with transcript line number citations

05Automated status updates for issue tracking files

0635 GitHub stars

사용 사례

01Debugging why an AI agent provided incorrect or hallucinated information during a session

02Auditing agent behavior to determine if instructions need refinement or clarification

03Identifying missing documentation or core knowledge gaps in an agent's training set

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add unsupervisedcom/deepwork investigate-issues

For use in Claude.ai and ChatGPT

Download Skill