Knowledge Gap Review

Run this after completing a task to identify issues in information retrieval and propose candidates for improving skills/rules.

Purpose

A skill for AI-assisted development that helps verbalize project-specific tacit knowledge and continuously improve the knowledge base.

When to Use

• During self-review after completing a task

• When you felt unsure during implementation

• When you had to repeatedly refer to existing code

Review Process

Step 1: Reflection Perspectives

Review the most recent task from the following perspectives:

Items that took time to research

• Information that required three or more tool calls to find

• Implicit patterns understood by referencing multiple parts of existing code

• Areas where you had to proceed by guessing because documentation was not found

Items where decisions were difficult

• Cases with multiple possible implementation approaches

• Situations where you checked existing patterns for error codes or naming conventions

• Parts where you were not confident whether your choice was correct

Information missing from existing skills/rules

• Cases where you referenced a skill but the required information was missing

• Mistakes or rework that could have been avoided if a rule had existed

Step 2: Classify the Issues

Classify the identified issues based on the following criteria:

Difficulty Criteria

High Took more than 10 minutes / required multiple files

Medium Took 5–10 minutes / referenced 1–2 existing files

Low Less than 5 minutes / easy to find but worth recording

Step 3: Propose Where to Maintain Them

Propose the appropriate destination based on the nature of each issue:

Destination Suitable Cases

.claude/skills/

Knowledge referenced during specific tasks (invoked by command)

.claude/rules/

Rules that should always be applied (auto-loaded)

CLAUDE.md

Project-wide principles and architecture

Update existing file Can be handled by extending an existing skill/rule

Output Format

Output the results using the following format:

Knowledge Gap Review Results

Identified Issues

Details

1. (Issue name)

Current Problem: (What made it difficult)

Required Information:

• (Specific information that was needed 1)
• (Specific information that was needed 2)

Recommended Action: (File name and a summary of what should be added)

Next Steps

• Create a GitHub issue (label: ai-knowledge-gap)
• Create or update skill/rule files

Example

/knowledge-gap

After execution, the AI reviews the most recent task and outputs the identified issues and proposals in the format above.

Related

• Existing skills: .claude/skills/

• Existing rules: .claude/rules/

• CLAUDE.md: CLAUDE.md

• GitHub label: ai-knowledge-gap (for tracking issues)