By name: semantic-knowledge-repository description: Structures investigation results into a persistent knowledge base (facts, inferences, keywords). Use when a task yields significant findings, code analysis, or research results that should be saved for future reuse, or when searching for previously stored knowledge.
Semantic Knowledge Repository (SKR)
You are equipped with the Semantic Knowledge Repository (SKR) skill. Your objective is to proactively manage and query a structured knowledge base located at ~/.agents/knowledges/ to ensure investigation results, architectural decisions, and bug resolutions are reusable across sessions.
Core Directives
- Proactive Saving: You MUST automatically extract and save knowledge without asking the user when you identify:
- A complex bug root cause.
- A newly discovered architectural pattern or convention in the codebase.
- Significant, non-trivial research results from the web or documentation.
- A sequence of difficult steps required to achieve a specific goal.
- Proactive Searching: When starting a new task, especially one involving unfamiliar codebase areas or complex requirements, you SHOULD search the repository first to see if relevant knowledge already exists.
Usage Instructions
1. Saving Knowledge
When you decide to save knowledge, you must format it into a specific JSON structure.
Required JSON Structure:
facts(Array of Strings): Objective data, paths, and proper nouns. No hallucinations.inferences(Array of Strings): Logical conclusions or implicit assumptions (separate from facts).keywords(Array of Strings): Tags and synonyms for searchability (use broad and specific terms).confidence_score(Integer): 1-100 score based on how reliable the information is.summary(String): A concise overview, easily readable at a glance.
(See references/schema.md for the full schema details if needed)
Execution:
Execute the following command using run_shell_command. Use kebab-case for the <filename> (e.g., oauth2-flow.json).
node <path_to_skill_directory>/scripts/save_knowledge.cjs <filename> '<json_content>'
Note: Replace <path_to_skill_directory> with the actual absolute path to where this skill is located.
Post-Action:
Notify the user after successfully saving the file: "Saved to Semantic Knowledge Repository: [filename].json"
2. Searching Knowledge
To find previously stored knowledge, execute the search script. Provide a clear, concise query (e.g., a specific framework name, bug type, or architectural concept).
Execution:
node <path_to_skill_directory>/scripts/search_knowledge.cjs <query>
Post-Action:
If the search returns relevant filenames, use the read_file tool to read those specific JSON files from ~/.agents/knowledges/ to incorporate their context into your current task.
Best Practices
- Accuracy over Speed: Ensure
factsare empirically verified. If you are unsure about an inference, lower theconfidence_score. - Escape Quotes: When passing
<json_content>to the bash command, ensure single quotes within the JSON are properly escaped or use an appropriate shell quoting strategy to prevent syntax errors.