case-feedback-skill

zai-org's avatarfrom zai-org

Run the case feedback script to submit feedback for the current conversation. Only use when invoked by case-feedback-agent.

⭐3stars🔀1forks📁View on GitHub🕐Updated Dec 25, 2025⬇️1downloads

When & Why to Use This Skill

This Claude skill facilitates the automated submission of detailed conversation feedback and diagnostic data. By capturing comprehensive context—including task objectives, execution logs, and error details—it enables developers to monitor agent performance, debug issues, and gather actionable insights to improve AI interactions and system reliability.

Use Cases

  • Bug Reporting: Automatically capture and submit error logs, code snippets, and conversation history when an agent encounters unexpected behavior.
  • Performance Monitoring: Collect structured feedback on the quality of AI-generated solutions to identify patterns in model performance and accuracy.
  • User Feedback Loop: Enable a seamless way for users to provide direct feedback on conversation helpfulness, bundled with technical metadata for developer review.
  • Incident Documentation: Summarize complex multi-step interactions and system states to provide clear, concise context for technical support or debugging teams.
namecase-feedback-skill
descriptionRun the case feedback script to submit feedback for the current conversation. Only use when invoked by case-feedback-agent.
allowed-toolsBash, Read

Case Feedback Skill

Execute the feedback submission script and return the result.

Critical constraint

Run the script exactly once — regardless of success or failure, execute it once and return the outcome.

Execution

Gather information

feedback:

  • If the user explicitly provided feedback text, use that directly
  • If the user describes a problem or issue, summarize it concisely as the feedback
  • Ask the user for clarification only if no feedback intent can be inferred

context:

The context contains a summary of the conversation, and must append the original, complete conversation history data.

Summarize the current conversation context, including:

  • What task the user was trying to accomplish
  • What operations were performed
  • Any errors or unexpected behaviors encountered
  • Relevant code snippets or file paths (keep it concise)

code_type:

Identify the programming language or code type involved (e.g., JavaScript, Python, Java). If not relevant, leave it blank.

request_id:

Extract the unique request ID or the session ID associated with this conversation or case. If not available, leave it blank.

happened_time:

Extract the timestamp when the issue occurred. If not mentioned, leave it blank.

Run the submission

Use Node.js to execute the bundled script, pay attention to the path changes in the Windows:

node scripts/submit-feedback.mjs --feedback "user feedback content" --context "conversation context summary" --code_type "the current code type, eg: javascript, typescript, python, java, etc. Not required." --happened_time "the time when the issue happened, eg: 2025-12-10 11:15:00. Not required." --request_id "the unique request id if available. Not required."

If your working directory is elsewhere, cd into the plugin root first or use an absolute path: node /absolute/path/to/glm-plan-bug/skills/case-feedback-skill/scripts/submit-feedback.mjs --feedback "..." --context "..."

Return the result

After execution, return the result to the caller:

  • Success: display the submission confirmation
  • Failure: show the error details and likely cause