PQL: Add question-type classification #20
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🚀 PromptQL Build CompleteBuild Version: Description: PR #20: PQL: Add question-type classification |
|
No description provided. |
✅ PromptQL Build AppliedBuild Version: |
robertjdominguez
added a commit
that referenced
this pull request
Aug 10, 2025
## Description Updated the DocsBot system instructions to be more precise about validation requirements and question handling. ### Key Changes - **Added GraphQL exclusion rule** - Explicitly tells the bot to avoid GraphQL API details since PromptQL users don't need them - **Restructured validation workflow** - Replaced the generic embedding search approach with question-type-specific validation paths - **Added mandatory validation checkpoints** - CLI and metadata questions now require exact documentation page validation before responding - **Strengthened no-guessing policy** - Zero tolerance for inventing commands or examples not found in documentation Previously, the bot used a general embedding search approach that could lead to hallucinated responses: ````yaml path=pql/globals/metadata/promptql-config.hml mode=EXCERPT ## Response Workflow 1. Transform user query into embedding 2. Find 3-5 most relevant chunks 3. **For CLI questions**: Validate commands exist in documentation before responding ```` Now it enforces strict validation based on question classification: ````yaml path=pql/globals/metadata/promptql-config.hml mode=EXCERPT ## Question Classification Before starting the response workflow, classify the question type: - **CLI question**: Contains words like "command", "ddn", "init", "introspect", "build", etc. - **Metadata question**: Mentions "metadata", "configuration", "YAML", specific objects like "models", "permissions", etc. - **General question**: Everything else ## Response Workflow 1. **Classify question type** (CLI/Metadata/General) 2. **Mandatory validation**: - CLI: Query exact command documentation page - Metadata: Query exact metadata object page - General: Transform query and search embeddings ```` This ensures the bot can't provide CLI commands or metadata examples unless it finds the exact documentation page first. The validation checkpoints act as hard stops - if a command or metadata object isn't documented, the bot admits it rather than guessing.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Updated the DocsBot system instructions to be more precise about validation requirements and question handling.
Key Changes
Previously, the bot used a general embedding search approach that could lead to hallucinated responses:
Now it enforces strict validation based on question classification:
This ensures the bot can't provide CLI commands or metadata examples unless it finds the exact documentation page first. The validation checkpoints act as hard stops - if a command or metadata object isn't documented, the bot admits it rather than guessing.