This file is the bridge entry point. The actual capability definitions are in the internal/ directory.
Routing Table
After identifying the user's intent, read the corresponding pipeline file and execute according to its instructions.
| Intent | Trigger Words | Read File |
|---|---|---|
create |
Create a knowledge network, create a new BKN, model based on a PRD | internal/_pipelines/create.md |
extract |
Extract object classes, extract entity relationships from documents | internal/_pipelines/extract.md |
read |
Query, search, list, view | internal/_pipelines/read.md |
update |
Modify, edit, update, rebind | internal/_pipelines/update.md |
delete |
Delete, clean up, remove | internal/_pipelines/delete.md |
copy |
Copy, clone, generate a new version based on an existing one | internal/_pipelines/copy.md |
validate |
Check the network, evaluate quality, diagnose issues, verify integrity | internal/_pipelines/validate.md |
skill_generate |
Add skills, extract into standalone skills, fill capability gaps | internal/_pipelines/skill-gen.md |
feedback_review |
Inspection feedback, usage issues, scheduled task triggers, improvement summaries, trace analysis results | internal/_pipelines/feedback.md |
Execution Protocol
- Identify intent → Output: identified intent + routing target + input summary + confirmation request
- During the routing confirmation stage, only display high-level information, not object/relationship details
- After user confirmation → Readthe pipeline file in the table above, and execute step by step according to the instructions in the file
- For each skill referenced in the pipeline, read its SKILL.md according to the relative path indicated in the pipeline
Shared Conventions
Before execution, first read:internal/_shared/contract.md
Constraints
- When unconfirmed, only repeat "route identification + confirmation request"
- Switching between processes must go through routing confirmation again