Current file: app/src/lib/tools/readme-generator.ts
Current model: llama-3.3-70b
Current approach: Single prompt asking for a full README. No structure validation, no badge verification, no section completeness check.
Problems with current approach:
- Generated shields.io badge URLs may be broken or use incorrect syntax.
- Installation commands may not match the described tech stack.
- No validation that all standard README sections are present.
- Code examples may have syntax errors.
Upgrade plan:
| Step |
Agent |
Action |
| 1 |
Project Analyzer |
Analyze the project description, tech stack, and features to build a structured metadata object: language, package manager, framework, entry points, dependencies. |
| 2 |
Section Planner |
Programmatic: Define which sections are required based on project type. Generate a section checklist (Title, Badges, Description, Features, Quick Start, Usage, API Reference, Config, Contributing, License). |
| 3 |
README Writer |
Generate the full README content following the section plan and project metadata. |
| 4 |
Section Validator |
Programmatic: Verify all required sections are present. Validate shields.io badge URLs match expected format. Check that code blocks have language identifiers. |
| 5 |
Refinement Agent |
If any sections are missing or badges are malformed, regenerate only those sections. Max 2 retries. |
- You are free to enhance the agents stacks in the above plan layout, the above one is just for reference. You can enhance more if needed.
Model suggestions to start with:
- Step 1: Try
deepseek-v3.2 for project analysis (lightweight).
- Steps 3 and 5: Try
llama-3.3-70b (current model, good at technical writing). Also try qwen-3-32b or kimi-k2.6 and compare.
Model Selection Guidance
- You are free to pick any model from the Oxlo catalog based on your own testing and evaluation.
- The Models suggestions above, not mandates. Try them first, and if they do not meet the accuracy target, experiment with alternatives.
Compare against: Claude Sonnet 4.6 Thinking (excellent at technical documentation).
Acceptance criteria:
- All required sections present in 100% of generated READMEs.
- Shields.io badge URLs are syntactically valid.
- Installation commands match the described tech stack.
- Overall quality matches or exceeds Claude Sonnet 4.6 Thinking on test cases.
- Overall accuracy at 80%+.
Current file:
app/src/lib/tools/readme-generator.tsCurrent model:
llama-3.3-70bCurrent approach: Single prompt asking for a full README. No structure validation, no badge verification, no section completeness check.
Problems with current approach:
Upgrade plan:
Model suggestions to start with:
deepseek-v3.2for project analysis (lightweight).llama-3.3-70b(current model, good at technical writing). Also tryqwen-3-32borkimi-k2.6and compare.Model Selection Guidance
Compare against: Claude Sonnet 4.6 Thinking (excellent at technical documentation).
Acceptance criteria: